d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b9c473b2ed
freebsd-nq/sys/kern/subr_witness.c

2841 lines
77 KiB
C
Raw Normal View History

Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*-
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* Copyright (c) 2008 Isilon Systems, Inc.
* Copyright (c) 2008 Ilya Maykov <ivmaykov@gmail.com>
* Copyright (c) 1998 Berkeley Software Design, Inc.
* All rights reserved.
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Berkeley Software Design Inc's name may not be used to endorse or
* promote products derived from this software without specific prior
* written permission.
*
* THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
* and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
*/
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
/*
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
* Implementation of the `witness' lock verifier. Originally implemented for
* mutexes in BSD/OS. Extended to handle generic lock objects and lock
* classes in FreeBSD.
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
*/
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*
* Main Entry: witness
* Pronunciation: 'wit-n&s
* Function: noun
* Etymology: Middle English witnesse, from Old English witnes knowledge,
* testimony, witness, from 2wit
* Date: before 12th century
* 1 : attestation of a fact or event : TESTIMONY
* 2 : one that gives evidence; specifically : one who testifies in
* a cause or before a judicial tribunal
* 3 : one asked to be present at a transaction so as to be able to
* testify to its having taken place
* 4 : one who has personal knowledge of something
* 5 a : something serving as evidence or proof : SIGN
* b : public affirmation by word or example of usually
* religious faith or conviction <the heroic witness to divine
* life -- Pilot>
* 6 capitalized : a member of the Jehovah's Witnesses
*/
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
/*
* Special rules concerning Giant and lock orders:
*
* 1) Giant must be acquired before any other mutexes. Stated another way,
* no other mutex may be held when Giant is acquired.
*
* 2) Giant must be released when blocking on a sleepable lock.
*
* This rule is less obvious, but is a result of Giant providing the same
* semantics as spl(). Basically, when a thread sleeps, it must release
* Giant. When a thread blocks on a sleepable lock, it sleeps. Hence rule
* 2).
*
* 3) Giant may be acquired before or after sleepable locks.
*
* This rule is also not quite as obvious. Giant may be acquired after
* a sleepable lock because it is a non-sleepable lock and non-sleepable
* locks may always be acquired while holding a sleepable lock. The second
* case, Giant before a sleepable lock, follows from rule 2) above. Suppose
* you have two threads T1 and T2 and a sleepable lock X. Suppose that T1
* acquires X and blocks on Giant. Then suppose that T2 acquires Giant and
* blocks on X. When T2 blocks on X, T2 will release Giant allowing T1 to
* execute. Thus, acquiring Giant both before and after a sleepable lock
* will not result in a lock order reversal.
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Include the right headers to get the DDB #define and the db_active variable.
2000-11-15 22:08:16 +00:00
#include "opt_ddb.h"
Fix witness(4) warnings about mutex use. Group mutexes used in hwpmc(4) into 3 "types" in the sense of witness(4): - leaf spin mutexes---only one of these should be held at a time, so these mutexes are specified as belonging to a single witness type "pmc-leaf". - `struct pmc_owner' descriptors are protected by a spin mutex of witness type "pmc-owner-proc". Since we call wakeup_one() while holding these mutexes, the witness type of these mutexes needs to dominate that of "sleepq chain" mutexes. - logger threads use a sleep mutex, of type "pmc-sleep". Submitted by: wkoszek (earlier patch)
2007-04-19 08:02:51 +00:00
#include "opt_hwpmc_hooks.h"
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#include "opt_stack.h"
- Use MUTEX_DECLARE() and MTX_COLD for the WITNESS code's internal mutex so it can function before malloc(9) is up and running. - Add two new options WITNESS_DDB and WITNESS_SKIPSPIN. If WITNESS_SKIPSPIN is enabled, then spin mutexes are ignored by the WITNESS code. If WITNESS_DDB is turned on and DDB is compiled into the kernel, then the kernel will drop into DDB when either a lock hierarchy violation occurs or mutexes are held when going to sleep. - Add some new sysctls: debug.witness_ddb is a read-write sysctl that corresponds to WITNESS_DDB. The kernel option merely changes the default value to on at boot. debug.witness_skipspin is a read-only sysctl that one can use to determine if the kernel was compiled with WITNESS_SKIPSPIN. - Wipe out the BSD/OS-specific lock order lists. We get to build our own lists now as we add mutexes to the kernel.
2000-10-27 02:59:30 +00:00
#include "opt_witness.h"
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#include <sys/param.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
#include <sys/bus.h>
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
#include <sys/kdb.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
#include <sys/kernel.h>
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#include <sys/ktr.h>
#include <sys/lock.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
#include <sys/malloc.h>
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#include <sys/mutex.h>
Sweep kernel replacing suser(9) calls with priv(9) calls, assigning specific privilege names to a broad range of privileges. These may require some future tweaking. Sponsored by: nCircle Network Security, Inc. Obtained from: TrustedBSD Project Discussed on: arch@ Reviewed (at least in part) by: mlaier, jmg, pjd, bde, ceri, Alex Lyashkov <umka at sevcity dot net>, Skip Ford <skip dot ford at verizon dot net>, Antoine Brodin <antoine dot brodin at laposte dot net>
2006-11-06 13:42:10 +00:00
#include <sys/priv.h>
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#include <sys/proc.h>
Add a new witness sysctl which returns the relations between any lock and its children in the form: "parent","child" so that head and bottom of an oriented graph can be easilly detected and various form of diagrams can be build. The sysctl is called debug.witness.graphs and it is read-only; in order to get the list of relations, a simple: #sysctl debug.witness.graphs will do the trick. This approach has been choosen in order to support easilly things like the DOT format and such. Soon, an auto-explicative awk script, which filters simple informations returned by the sysctl and converts them into a real DOT script, will be committed to the repository between examples. Discussed with: rwatson
2008-05-07 21:41:36 +00:00
#include <sys/sbuf.h>
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
#include <sys/sched.h>
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#include <sys/stack.h>
- Use MUTEX_DECLARE() and MTX_COLD for the WITNESS code's internal mutex so it can function before malloc(9) is up and running. - Add two new options WITNESS_DDB and WITNESS_SKIPSPIN. If WITNESS_SKIPSPIN is enabled, then spin mutexes are ignored by the WITNESS code. If WITNESS_DDB is turned on and DDB is compiled into the kernel, then the kernel will drop into DDB when either a lock hierarchy violation occurs or mutexes are held when going to sleep. - Add some new sysctls: debug.witness_ddb is a read-write sysctl that corresponds to WITNESS_DDB. The kernel option merely changes the default value to on at boot. debug.witness_skipspin is a read-only sysctl that one can use to determine if the kernel was compiled with WITNESS_SKIPSPIN. - Wipe out the BSD/OS-specific lock order lists. We get to build our own lists now as we add mutexes to the kernel.
2000-10-27 02:59:30 +00:00
#include <sys/sysctl.h>
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#include <sys/systm.h>
- Make the mutex code almost completely machine independent. This greatly reducues the maintenance load for the mutex code. The only MD portions of the mutex code are in machine/mutex.h now, which include the assembly macros for handling mutexes as well as optionally overriding the mutex micro-operations. For example, we use optimized micro-ops on the x86 platform #ifndef I386_CPU. - Change the behavior of the SMP_DEBUG kernel option. In the new code, mtx_assert() only depends on INVARIANTS, allowing other kernel developers to have working mutex assertiions without having to include all of the mutex debugging code. The SMP_DEBUG kernel option has been renamed to MUTEX_DEBUG and now just controls extra mutex debugging code. - Abolish the ugly mtx_f hack. Instead, we dynamically allocate seperate mtx_debug structures on the fly in mtx_init, except for mutexes that are initiated very early in the boot process. These mutexes are declared using a special MUTEX_DECLARE() macro, and use a new flag MTX_COLD when calling mtx_init. This is still somewhat hackish, but it is less evil than the mtx_f filler struct, and the mtx struct is now the same size with and without mutex debugging code. - Add some micro-micro-operation macros for doing the actual atomic operations on the mutex mtx_lock field to make it easier for other archs to override/optimize mutex ops if needed. These new tiny ops also clean up the code in some places by replacing long atomic operation function calls that spanned 2-3 lines with a short 1-line macro call. - Don't call mi_switch() from mtx_enter_hard() when we block while trying to obtain a sleep mutex. Calling mi_switch() would bogusly release Giant before switching to the next process. Instead, inline most of the code from mi_switch() in the mtx_enter_hard() function. Note that when we finally kill Giant we can back this out and go back to calling mi_switch().
2000-10-20 07:26:37 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#ifdef DDB
Include the right headers to get the DDB #define and the db_active variable.
2000-11-15 22:08:16 +00:00
#include <ddb/ddb.h>
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#endif
Include the right headers to get the DDB #define and the db_active variable.
2000-11-15 22:08:16 +00:00
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
#include <machine/stdarg.h>
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#if !defined(DDB) && !defined(STACK)
#error "DDB or STACK options are required for WITNESS"
#endif
Axe KTR_ALQ_MASK now that KTR_WITNESS is off unless you hack an #ifdef in subr_witness.c. I did add a comment in subr_witness.c noting that KTR_WITNESS is incompatible with KTR_ALQ.
2006-01-25 14:57:23 +00:00
/* Note that these traces do not work with KTR_ALQ. */
- Add a new KTR_SUBSYS in place of KTR_SPARE1 to serve as a subsystem placeholder similar to KTR_DEV. Explain the use of KTR_DEV and KTR_SUBSYS in a comment as well. - Retire KTR_WITNESS and instead have KTR_WITNESS default to off but use KTR_SUBSYS if it is enabled.
2006-01-24 22:23:45 +00:00
#if 0
#define KTR_WITNESS KTR_SUBSYS
#else
#define KTR_WITNESS 0
#endif
struct lock_instance and struct lock_list_entry don't need to be in the public namespace for WITNESS as they are only used internally so just move them in the private namespace for the subsystem (with all related supporting definitions).
2008-04-13 01:20:47 +00:00
#define LI_RECURSEMASK 0x0000ffff /* Recursion depth of lock instance. */
#define LI_EXCLUSIVE 0x00010000 /* Exclusive lock instance. */
Add functions WITNESS so it can be asserted that the lock is not released for a section of code, this uses WITNESS_NORELEASE() and WITNESS_RELEASEOK() to mark the boundaries. Both functions require the lock to be held when calling. This is intended for scenarios like a bus asserting that the bus lock is not dropped during a driver call. There doesn't appear to be a man page to document this in. Reviewed by: jhb
2009-01-21 04:19:18 +00:00
#define LI_NORELEASE 0x00020000 /* Lock not allowed to be released. */
struct lock_instance and struct lock_list_entry don't need to be in the public namespace for WITNESS as they are only used internally so just move them in the private namespace for the subsystem (with all related supporting definitions).
2008-04-13 01:20:47 +00:00
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
/* Define this to check for blessed mutexes */
#undef BLESSING
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#define WITNESS_COUNT 1024
#define WITNESS_CHILDCOUNT (WITNESS_COUNT * 4)
#define WITNESS_HASH_SIZE 251 /* Prime, gives load factor < 2 */
Bump the witness pendlist to 768 to accomodate the increased number of spinlocks.
2010-07-29 16:13:26 +00:00
#define WITNESS_PENDLIST 768
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* Allocate 256 KB of stack data space */
#define WITNESS_LO_DATA_COUNT 2048
/* Prime, gives load factor of ~2 at full load */
#define WITNESS_LO_HASH_SIZE 1021
- Don't use a union and fun tricks to shave one extra pointer off of struct mtx right now as it makes debugging harder. When we are in optimizing mode, we can revisit this. - Fix the KTR trace messages to use %p rather than 0x%p to avoid duplicate 0x's in KTR output. - During witness_fixup, release Giant so that witness doesn't get confused. Also, grab all_mtx while walking the list of mutexes. - Remove w_sleep and w_recurse. Instead, perform checks on mutexes using the mutex's mtx_flags field. - Allow debug.witness_ddb and debug.witness_skipspin to be set from the loader. - Add Giant to the front of existing order_list entries to help ensure Giant is always first. - Add an order entry for the various proc locks. Note that this only helps keep proc in order mostly as the allproc and proctree mutexes are only obtained during a lockmgr operation on the specified mutex.
2001-01-24 10:57:01 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* XXX: This is somewhat bogus, as we assume here that at most 2048 threads
* will hold LOCK_NCHILDREN locks. We handle failure ok, and we should
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
* probably be safe for the most part, but it's still a SWAG.
- Don't use a union and fun tricks to shave one extra pointer off of struct mtx right now as it makes debugging harder. When we are in optimizing mode, we can revisit this. - Fix the KTR trace messages to use %p rather than 0x%p to avoid duplicate 0x's in KTR output. - During witness_fixup, release Giant so that witness doesn't get confused. Also, grab all_mtx while walking the list of mutexes. - Remove w_sleep and w_recurse. Instead, perform checks on mutexes using the mutex's mtx_flags field. - Allow debug.witness_ddb and debug.witness_skipspin to be set from the loader. - Add Giant to the front of existing order_list entries to help ensure Giant is always first. - Add an order entry for the various proc locks. Note that this only helps keep proc in order mostly as the allproc and proctree mutexes are only obtained during a lockmgr operation on the specified mutex.
2001-01-24 10:57:01 +00:00
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#define LOCK_NCHILDREN 5
#define LOCK_CHILDCOUNT 2048
- Don't use a union and fun tricks to shave one extra pointer off of struct mtx right now as it makes debugging harder. When we are in optimizing mode, we can revisit this. - Fix the KTR trace messages to use %p rather than 0x%p to avoid duplicate 0x's in KTR output. - During witness_fixup, release Giant so that witness doesn't get confused. Also, grab all_mtx while walking the list of mutexes. - Remove w_sleep and w_recurse. Instead, perform checks on mutexes using the mutex's mtx_flags field. - Allow debug.witness_ddb and debug.witness_skipspin to be set from the loader. - Add Giant to the front of existing order_list entries to help ensure Giant is always first. - Add an order entry for the various proc locks. Note that this only helps keep proc in order mostly as the allproc and proctree mutexes are only obtained during a lockmgr operation on the specified mutex.
2001-01-24 10:57:01 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#define MAX_W_NAME 64
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#define BADSTACK_SBUF_SIZE (256 * WITNESS_COUNT)
Re-add r212370 now that the LOR in powerpc64 has been resolved: Add a drain function for struct sysctl_req, and use it for a variety of handlers, some of which had to do awkward things to get a large enough SBUF_FIXEDLEN buffer. Note that some sysctl handlers were explicitly outputting a trailing NUL byte. This behaviour was preserved, though it should not be necessary. Reviewed by: phk (original patch)
2010-09-16 16:13:12 +00:00
#define FULLGRAPH_SBUF_SIZE 512
struct lock_instance and struct lock_list_entry don't need to be in the public namespace for WITNESS as they are only used internally so just move them in the private namespace for the subsystem (with all related supporting definitions).
2008-04-13 01:20:47 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* These flags go in the witness relationship matrix and describe the
* relationship between any two struct witness objects.
*/
#define WITNESS_UNRELATED 0x00 /* No lock order relation. */
#define WITNESS_PARENT 0x01 /* Parent, aka direct ancestor. */
#define WITNESS_ANCESTOR 0x02 /* Direct or indirect ancestor. */
#define WITNESS_CHILD 0x04 /* Child, aka direct descendant. */
#define WITNESS_DESCENDANT 0x08 /* Direct or indirect descendant. */
#define WITNESS_ANCESTOR_MASK (WITNESS_PARENT | WITNESS_ANCESTOR)
#define WITNESS_DESCENDANT_MASK (WITNESS_CHILD | WITNESS_DESCENDANT)
#define WITNESS_RELATED_MASK \
(WITNESS_ANCESTOR_MASK | WITNESS_DESCENDANT_MASK)
#define WITNESS_REVERSAL 0x10 /* A lock order reversal has been
* observed. */
#define WITNESS_RESERVED1 0x20 /* Unused flag, reserved. */
#define WITNESS_RESERVED2 0x40 /* Unused flag, reserved. */
#define WITNESS_LOCK_ORDER_KNOWN 0x80 /* This lock order is known. */
/* Descendant to ancestor flags */
#define WITNESS_DTOA(x) (((x) & WITNESS_RELATED_MASK) >> 2)
/* Ancestor to descendant flags */
#define WITNESS_ATOD(x) (((x) & WITNESS_RELATED_MASK) << 2)
#define WITNESS_INDEX_ASSERT(i) \
MPASS((i) > 0 && (i) <= w_max_used_index && (i) < WITNESS_COUNT)
Mark MALLOC_DEFINEs static that have no corresponding MALLOC_DECLAREs. This means that their use is restricted to a single C file.
2011-11-07 06:44:47 +00:00
static MALLOC_DEFINE(M_WITNESS, "Witness", "Witness");
Remove MUTEX_DECLARE() and MTX_COLD. Instead, postpone full mutex initialization until after malloc() is safe to call, then iterate through all mutexes and complete their initialization. This change is necessary in order to avoid some circular bootstrapping dependencies.
2001-01-21 07:52:20 +00:00
struct lock_instance and struct lock_list_entry don't need to be in the public namespace for WITNESS as they are only used internally so just move them in the private namespace for the subsystem (with all related supporting definitions).
2008-04-13 01:20:47 +00:00
/*
* Lock instances. A lock instance is the data associated with a lock while
* it is held by witness. For example, a lock instance will hold the
* recursion count of a lock. Lock instances are held in lists. Spin locks
* are held in a per-cpu list while sleep locks are held in per-thread list.
*/
struct lock_instance {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct lock_object *li_lock;
const char *li_file;
int li_line;
u_int li_flags;
struct lock_instance and struct lock_list_entry don't need to be in the public namespace for WITNESS as they are only used internally so just move them in the private namespace for the subsystem (with all related supporting definitions).
2008-04-13 01:20:47 +00:00
};
/*
* A simple list type used to build the list of locks held by a thread
* or CPU. We can't simply embed the list in struct lock_object since a
* lock may be held by more than one thread if it is a shared lock. Locks
* are added to the head of the list, so we fill up each list entry from
* "the back" logically. To ease some of the arithmetic, we actually fill
* in each list entry the normal way (children[0] then children[1], etc.) but
* when we traverse the list we read children[count-1] as the first entry
* down to children[0] as the final entry.
*/
struct lock_list_entry {
struct lock_list_entry *ll_next;
struct lock_instance ll_children[LOCK_NCHILDREN];
u_int ll_count;
};
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* The main witness structure. One of these per named lock type in the system
* (for example, "vnode interlock").
*/
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct witness {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
char w_name[MAX_W_NAME];
uint32_t w_index; /* Index in the relationship matrix */
struct lock_class *w_class;
STAILQ_ENTRY(witness) w_list; /* List of all witnesses. */
STAILQ_ENTRY(witness) w_typelist; /* Witnesses of a type. */
struct witness *w_hash_next; /* Linked list in hash buckets. */
const char *w_file; /* File where last acquired */
uint32_t w_line; /* Line where last acquired */
uint32_t w_refcount;
uint16_t w_num_ancestors; /* direct/indirect
* ancestor count */
uint16_t w_num_descendants; /* direct/indirect
* descendant count */
int16_t w_ddb_level;
o Use unsigned for bit fields. o Use NULL for pointers in preference to 0.
2009-02-03 07:50:41 +00:00
unsigned w_displayed:1;
unsigned w_reversed:1;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
};
Remove MUTEX_DECLARE() and MTX_COLD. Instead, postpone full mutex initialization until after malloc() is safe to call, then iterate through all mutexes and complete their initialization. This change is necessary in order to avoid some circular bootstrapping dependencies.
2001-01-21 07:52:20 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
STAILQ_HEAD(witness_list, witness);
/*
* The witness hash table. Keys are witness names (const char *), elements are
* witness objects (struct witness *).
*/
struct witness_hash {
struct witness *wh_array[WITNESS_HASH_SIZE];
uint32_t wh_size;
uint32_t wh_count;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
};
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Key type for the lock order data hash table.
*/
struct witness_lock_order_key {
uint16_t from;
uint16_t to;
};
struct witness_lock_order_data {
struct stack wlod_stack;
struct witness_lock_order_key wlod_key;
struct witness_lock_order_data *wlod_next;
};
/*
* The witness lock order data hash table. Keys are witness index tuples
* (struct witness_lock_order_key), elements are lock order data objects
* (struct witness_lock_order_data).
*/
struct witness_lock_order_hash {
struct witness_lock_order_data *wloh_array[WITNESS_LO_HASH_SIZE];
u_int wloh_size;
u_int wloh_count;
};
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#ifdef BLESSING
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct witness_blessed {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
const char *b_lock1;
const char *b_lock2;
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
};
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#endif
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
struct witness_pendhelp {
const char *wh_type;
struct lock_object *wh_lock;
};
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness_order_list_entry {
const char *w_name;
struct lock_class *w_class;
};
/*
* Returns 0 if one of the locks is a spin lock and the other is not.
* Returns 1 otherwise.
*/
static __inline int
witness_lock_type_equal(struct witness *w1, struct witness *w2)
{
return ((w1->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)) ==
(w2->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)));
}
static __inline int
witness_lock_order_key_empty(const struct witness_lock_order_key *key)
{
return (key->from == 0 && key->to == 0);
}
static __inline int
witness_lock_order_key_equal(const struct witness_lock_order_key *a,
const struct witness_lock_order_key *b)
{
return (a->from == b->from && a->to == b->to);
}
static int _isitmyx(struct witness *w1, struct witness *w2, int rmask,
const char *fname);
#ifdef KDB
static void _witness_debugger(int cond, const char *msg);
#endif
static void adopt(struct witness *parent, struct witness *child);
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
#ifdef BLESSING
static int blessed(struct witness *, struct witness *);
#endif
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
static void depart(struct witness *w);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static struct witness *enroll(const char *description,
struct lock_class *lock_class);
static struct lock_instance *find_instance(struct lock_list_entry *list,
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
const struct lock_object *lock);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static int isitmychild(struct witness *parent, struct witness *child);
static int isitmydescendant(struct witness *parent, struct witness *child);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static void itismychild(struct witness *parent, struct witness *child);
static int sysctl_debug_witness_badstacks(SYSCTL_HANDLER_ARGS);
- Remove witness_dead and just use witness_watch instead. If witness_watch is set to 0, it now has the same affect as setting witness_dead used to have. - Added a sysctl handler that allows root to change witness_watch from a non-zero value to zero to disable witness at runtime. Note that you can't turn witness back on once it is off. You can only turn it off as a one-way switch. - Added a comment describing the possible values of witness_watch.
2003-03-24 21:03:53 +00:00
static int sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static int sysctl_debug_witness_fullgraph(SYSCTL_HANDLER_ARGS);
static void witness_add_fullgraph(struct sbuf *sb, struct witness *parent);
#ifdef DDB
static void witness_ddb_compute_levels(void);
There is not a good reason to have a different prototype for db_printf() when compared to printf(). Unify it by returning the number of characters displayed for db_printf() as well. MFC after: 7 days
2010-05-11 17:01:14 +00:00
static void witness_ddb_display(int(*)(const char *fmt, ...));
static void witness_ddb_display_descendants(int(*)(const char *fmt, ...),
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness *, int indent);
There is not a good reason to have a different prototype for db_printf() when compared to printf(). Unify it by returning the number of characters displayed for db_printf() as well. MFC after: 7 days
2010-05-11 17:01:14 +00:00
static void witness_ddb_display_list(int(*prnt)(const char *fmt, ...),
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness_list *list);
static void witness_ddb_level_descendants(struct witness *parent, int l);
static void witness_ddb_list(struct thread *td);
#endif
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static void witness_free(struct witness *m);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static struct witness *witness_get(void);
static uint32_t witness_hash_djb2(const uint8_t *key, uint32_t size);
static struct witness *witness_hash_get(const char *key);
static void witness_hash_put(struct witness *w);
static void witness_init_hash_tables(void);
static void witness_increment_graph_generation(void);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static void witness_lock_list_free(struct lock_list_entry *lle);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static struct lock_list_entry *witness_lock_list_get(void);
static int witness_lock_order_add(struct witness *parent,
struct witness *child);
static int witness_lock_order_check(struct witness *parent,
struct witness *child);
static struct witness_lock_order_data *witness_lock_order_get(
struct witness *parent,
struct witness *child);
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
static void witness_list_lock(struct lock_instance *instance,
int (*prnt)(const char *fmt, ...));
Add functions WITNESS so it can be asserted that the lock is not released for a section of code, this uses WITNESS_NORELEASE() and WITNESS_RELEASEOK() to mark the boundaries. Both functions require the lock to be held when calling. This is intended for scenarios like a bus asserting that the bus lock is not dropped during a driver call. There doesn't appear to be a man page to document this in. Reviewed by: jhb
2009-01-21 04:19:18 +00:00
static void witness_setflag(struct lock_object *lock, int flag, int set);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
#ifdef KDB
#define witness_debugger(c) _witness_debugger(c, __func__)
#else
#define witness_debugger(c)
Silence compiler warnings when DDB is not defined. PR: 36002 Submitted by: Yoshikazu GOTO <goto@snowy.to>
2002-07-15 02:03:17 +00:00
#endif
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_debug, OID_AUTO, witness, CTLFLAG_RW, NULL,
"Witness Locking");
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
- Remove witness_dead and just use witness_watch instead. If witness_watch is set to 0, it now has the same affect as setting witness_dead used to have. - Added a sysctl handler that allows root to change witness_watch from a non-zero value to zero to disable witness at runtime. Note that you can't turn witness back on once it is off. You can only turn it off as a one-way switch. - Added a comment describing the possible values of witness_watch.
2003-03-24 21:03:53 +00:00
/*
Update description of witness_watch.
2008-09-24 18:47:24 +00:00
* If set to 0, lock order checking is disabled. If set to -1,
* witness is completely disabled. Otherwise witness performs full
* lock order checking for all locks. At runtime, lock order checking
* may be toggled. However, witness cannot be reenabled once it is
* completely disabled.
- Remove witness_dead and just use witness_watch instead. If witness_watch is set to 0, it now has the same affect as setting witness_dead used to have. - Added a sysctl handler that allows root to change witness_watch from a non-zero value to zero to disable witness at runtime. Note that you can't turn witness back on once it is off. You can only turn it off as a one-way switch. - Added a comment describing the possible values of witness_watch.
2003-03-24 21:03:53 +00:00
*/
Make the TUNABLE_*() macros look and behave more consistantly like the SYSCTL_*() macros. TUNABLE_INT_DECL() was an odd name because it didn't actually declare the int, which is what the name suggests it would do.
2001-06-06 22:17:08 +00:00
static int witness_watch = 1;
make witness it's own sysctl branch instead of using _ to do this. I have left the old tunables in to give people a few days to transition their loader.conf and sysctl.conf's over to the new names.. MFC after: 5 days
2004-09-06 23:27:28 +00:00
TUNABLE_INT("debug.witness.watch", &witness_watch);
SYSCTL_PROC(_debug_witness, OID_AUTO, watch, CTLFLAG_RW | CTLTYPE_INT, NULL, 0,
- Remove witness_dead and just use witness_watch instead. If witness_watch is set to 0, it now has the same affect as setting witness_dead used to have. - Added a sysctl handler that allows root to change witness_watch from a non-zero value to zero to disable witness at runtime. Note that you can't turn witness back on once it is off. You can only turn it off as a one-way switch. - Added a comment describing the possible values of witness_watch.
2003-03-24 21:03:53 +00:00
sysctl_debug_witness_watch, "I", "witness is watching lock operations");
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
#ifdef KDB
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* When KDB is enabled and witness_kdb is 1, it will cause the system
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
* to drop into kdebug() when:
Spell hierarchy correctly in comments. Submitted by: Wojciech A. Koszek dunstan at freebsd dot czest dot pl
2005-10-24 15:57:27 +00:00
* - a lock hierarchy violation occurs
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
* - locks are held when going to sleep.
*/
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
#ifdef WITNESS_KDB
int witness_kdb = 1;
- Use MUTEX_DECLARE() and MTX_COLD for the WITNESS code's internal mutex so it can function before malloc(9) is up and running. - Add two new options WITNESS_DDB and WITNESS_SKIPSPIN. If WITNESS_SKIPSPIN is enabled, then spin mutexes are ignored by the WITNESS code. If WITNESS_DDB is turned on and DDB is compiled into the kernel, then the kernel will drop into DDB when either a lock hierarchy violation occurs or mutexes are held when going to sleep. - Add some new sysctls: debug.witness_ddb is a read-write sysctl that corresponds to WITNESS_DDB. The kernel option merely changes the default value to on at boot. debug.witness_skipspin is a read-only sysctl that one can use to determine if the kernel was compiled with WITNESS_SKIPSPIN. - Wipe out the BSD/OS-specific lock order lists. We get to build our own lists now as we add mutexes to the kernel.
2000-10-27 02:59:30 +00:00
#else
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
int witness_kdb = 0;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
#endif
make witness it's own sysctl branch instead of using _ to do this. I have left the old tunables in to give people a few days to transition their loader.conf and sysctl.conf's over to the new names.. MFC after: 5 days
2004-09-06 23:27:28 +00:00
TUNABLE_INT("debug.witness.kdb", &witness_kdb);
SYSCTL_INT(_debug_witness, OID_AUTO, kdb, CTLFLAG_RW, &witness_kdb, 0, "");
Add a 'debug.witness_trace' sysctl (and tunable) when DDB is present. This causes LOR and could-sleep messages to come with a stack trace.
2003-02-13 01:35:56 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* When KDB is enabled and witness_trace is 1, it will cause the system
Add a 'debug.witness_trace' sysctl (and tunable) when DDB is present. This causes LOR and could-sleep messages to come with a stack trace.
2003-02-13 01:35:56 +00:00
* to print a stack trace:
Spell hierarchy correctly in comments. Submitted by: Wojciech A. Koszek dunstan at freebsd dot czest dot pl
2005-10-24 15:57:27 +00:00
* - a lock hierarchy violation occurs
Add a 'debug.witness_trace' sysctl (and tunable) when DDB is present. This causes LOR and could-sleep messages to come with a stack trace.
2003-02-13 01:35:56 +00:00
* - locks are held when going to sleep.
*/
int witness_trace = 1;
make witness it's own sysctl branch instead of using _ to do this. I have left the old tunables in to give people a few days to transition their loader.conf and sysctl.conf's over to the new names.. MFC after: 5 days
2004-09-06 23:27:28 +00:00
TUNABLE_INT("debug.witness.trace", &witness_trace);
SYSCTL_INT(_debug_witness, OID_AUTO, trace, CTLFLAG_RW, &witness_trace, 0, "");
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
#endif /* KDB */
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Use MUTEX_DECLARE() and MTX_COLD for the WITNESS code's internal mutex so it can function before malloc(9) is up and running. - Add two new options WITNESS_DDB and WITNESS_SKIPSPIN. If WITNESS_SKIPSPIN is enabled, then spin mutexes are ignored by the WITNESS code. If WITNESS_DDB is turned on and DDB is compiled into the kernel, then the kernel will drop into DDB when either a lock hierarchy violation occurs or mutexes are held when going to sleep. - Add some new sysctls: debug.witness_ddb is a read-write sysctl that corresponds to WITNESS_DDB. The kernel option merely changes the default value to on at boot. debug.witness_skipspin is a read-only sysctl that one can use to determine if the kernel was compiled with WITNESS_SKIPSPIN. - Wipe out the BSD/OS-specific lock order lists. We get to build our own lists now as we add mutexes to the kernel.
2000-10-27 02:59:30 +00:00
#ifdef WITNESS_SKIPSPIN
Make the TUNABLE_*() macros look and behave more consistantly like the SYSCTL_*() macros. TUNABLE_INT_DECL() was an odd name because it didn't actually declare the int, which is what the name suggests it would do.
2001-06-06 22:17:08 +00:00
int witness_skipspin = 1;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#else
Make the TUNABLE_*() macros look and behave more consistantly like the SYSCTL_*() macros. TUNABLE_INT_DECL() was an odd name because it didn't actually declare the int, which is what the name suggests it would do.
2001-06-06 22:17:08 +00:00
int witness_skipspin = 0;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#endif
make witness it's own sysctl branch instead of using _ to do this. I have left the old tunables in to give people a few days to transition their loader.conf and sysctl.conf's over to the new names.. MFC after: 5 days
2004-09-06 23:27:28 +00:00
TUNABLE_INT("debug.witness.skipspin", &witness_skipspin);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
SYSCTL_INT(_debug_witness, OID_AUTO, skipspin, CTLFLAG_RDTUN, &witness_skipspin,
0, "");
/*
* Call this to print out the relations between locks.
*/
SYSCTL_PROC(_debug_witness, OID_AUTO, fullgraph, CTLTYPE_STRING | CTLFLAG_RD,
NULL, 0, sysctl_debug_witness_fullgraph, "A", "Show locks relation graphs");
/*
* Call this to print out the witness faulty stacks.
*/
SYSCTL_PROC(_debug_witness, OID_AUTO, badstacks, CTLTYPE_STRING | CTLFLAG_RD,
NULL, 0, sysctl_debug_witness_badstacks, "A", "Show bad witness stacks");
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct mtx w_mtx;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* w_list */
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct witness_list w_free = STAILQ_HEAD_INITIALIZER(w_free);
static struct witness_list w_all = STAILQ_HEAD_INITIALIZER(w_all);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* w_typelist */
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct witness_list w_spin = STAILQ_HEAD_INITIALIZER(w_spin);
static struct witness_list w_sleep = STAILQ_HEAD_INITIALIZER(w_sleep);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* lock list */
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct lock_list_entry *w_lock_list_free = NULL;
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
static struct witness_pendhelp pending_locks[WITNESS_PENDLIST];
static u_int pending_cnt;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static int w_free_cnt, w_spin_cnt, w_sleep_cnt;
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
SYSCTL_INT(_debug_witness, OID_AUTO, free_cnt, CTLFLAG_RD, &w_free_cnt, 0, "");
SYSCTL_INT(_debug_witness, OID_AUTO, spin_cnt, CTLFLAG_RD, &w_spin_cnt, 0, "");
SYSCTL_INT(_debug_witness, OID_AUTO, sleep_cnt, CTLFLAG_RD, &w_sleep_cnt, 0,
"");
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static struct witness *w_data;
static uint8_t w_rmatrix[WITNESS_COUNT+1][WITNESS_COUNT+1];
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct lock_list_entry w_locklistdata[LOCK_CHILDCOUNT];
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static struct witness_hash w_hash; /* The witness hash table. */
/* The lock order data hash */
static struct witness_lock_order_data w_lodata[WITNESS_LO_DATA_COUNT];
static struct witness_lock_order_data *w_lofree = NULL;
static struct witness_lock_order_hash w_lohash;
static int w_max_used_index = 0;
static unsigned int w_generation = 0;
Change w_notrunning and w_stillcold from pointer to array so that sizeof returns what is expected. PR: kern/138557 Discussed with: brucec@ MFC after: 1 month
2009-09-06 13:31:05 +00:00
static const char w_notrunning[] = "Witness not running\n";
static const char w_stillcold[] = "Witness is still cold\n";
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct witness_order_list_entry order_lists[] = {
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
/*
* sx locks
*/
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ "proctree", &lock_class_sx },
{ "allproc", &lock_class_sx },
allprison mutex was converted to sx(9) lock.
2007-04-05 23:32:32 +00:00
{ "allprison", &lock_class_sx },
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
{ NULL, NULL },
/*
* Various mutexes
*/
Oops, fix the double faults people were seeing with the recent changes to witness. Sleepable locks such as sx locks always come before all mutexes including Giant. However, the static lock order list placed Giant before the proctree and allproc sx locks. This resulted in witness creating a cycle in its lock order "tree" (real trees don't have cycles) leading to infinite recursion and eventually a double fault. To fix, put Giant after sx locks in the lock order list.
2003-03-06 17:25:06 +00:00
{ "Giant", &lock_class_mtx_sleep },
Enforce correct ordering of the filedesc structure and pipe mutex, because WITNESS can get the order wrong if it guesses based on first use. Reviewed by: jhb, alfred
2002-12-22 16:32:34 +00:00
{ "pipe mutex", &lock_class_mtx_sleep },
Make funsetown() take a 'struct sigio **' so that the locking can be done internally. Ensure that no one can fsetown() to a dying process/pgrp. We need to check the process for P_WEXIT to see if it's exiting. Process groups are already safe because there is no such thing as a pgrp zombie, therefore the proctree lock completely protects the pgrp from having sigio structures associated with it after it runs funsetownlst. Add sigio lock to witness list under proctree and allproc, but over proc and pgrp. Seigo Tanimura helped with this.
2002-05-06 19:31:28 +00:00
{ "sigio lock", &lock_class_mtx_sleep },
Lock struct pgrp, session and sigio. New locks are: - pgrpsess_lock which locks the whole pgrps and sessions, - pg_mtx which protects the pgrp members, and - s_mtx which protects the session members. Please refer to sys/proc.h for the coverage of these locks. Changes on the pgrp/session interface: - pgfind() needs the pgrpsess_lock held. - The caller of enterpgrp() is responsible to allocate a new pgrp and session. - Call enterthispgrp() in order to enter an existing pgrp. - pgsignal() requires a pgrp lock held. Reviewed by: jhb, alfred Tested on: cvsup.jp.FreeBSD.org (which is a quad-CPU machine running -current)
2002-02-23 11:12:57 +00:00
{ "process group", &lock_class_mtx_sleep },
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ "process lock", &lock_class_mtx_sleep },
Lock struct pgrp, session and sigio. New locks are: - pgrpsess_lock which locks the whole pgrps and sessions, - pg_mtx which protects the pgrp members, and - s_mtx which protects the session members. Please refer to sys/proc.h for the coverage of these locks. Changes on the pgrp/session interface: - pgfind() needs the pgrpsess_lock held. - The caller of enterpgrp() is responsible to allocate a new pgrp and session. - Call enterthispgrp() in order to enter an existing pgrp. - pgsignal() requires a pgrp lock held. Reviewed by: jhb, alfred Tested on: cvsup.jp.FreeBSD.org (which is a quad-CPU machine running -current)
2002-02-23 11:12:57 +00:00
{ "session", &lock_class_mtx_sleep },
- There is no more "uidinfo struct" mutex. - The "uidinfo hash" lock is now a rwlock. Reminded by: kib
2008-03-17 11:48:40 +00:00
{ "uidinfo hash", &lock_class_rw },
Fix witness(4) warnings about mutex use. Group mutexes used in hwpmc(4) into 3 "types" in the sense of witness(4): - leaf spin mutexes---only one of these should be held at a time, so these mutexes are specified as belonging to a single witness type "pmc-leaf". - `struct pmc_owner' descriptors are protected by a spin mutex of witness type "pmc-owner-proc". Since we call wakeup_one() while holding these mutexes, the witness type of these mutexes needs to dominate that of "sleepq chain" mutexes. - logger threads use a sleep mutex, of type "pmc-sleep". Submitted by: wkoszek (earlier patch)
2007-04-19 08:02:51 +00:00
#ifdef HWPMC_HOOKS
{ "pmc-sleep", &lock_class_mtx_sleep },
#endif
"time lock" is no longer a spin-lock since r209371. Reported by: kib@
2010-06-21 21:15:51 +00:00
{ "time lock", &lock_class_mtx_sleep },
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ NULL, NULL },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
/*
* Sockets
*/
{ "accept", &lock_class_mtx_sleep },
Introduce socket and UNIX domain socket locks into hard-coded lock order definition for witness. Send lock before receive lock, and socket locks after accept but before select: filedesc -> accept -> so_snd -> so_rcv -> sellck All routing locks after send lock: so_rcv -> radix node head All protocol locks before socket locks: unp -> so_snd udp -> udpinp -> so_snd tcp -> tcpinp -> so_snd
2004-06-13 00:23:03 +00:00
{ "so_snd", &lock_class_mtx_sleep },
{ "so_rcv", &lock_class_mtx_sleep },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
{ "sellck", &lock_class_mtx_sleep },
{ NULL, NULL },
/*
* Routing
*/
Introduce socket and UNIX domain socket locks into hard-coded lock order definition for witness. Send lock before receive lock, and socket locks after accept but before select: filedesc -> accept -> so_snd -> so_rcv -> sellck All routing locks after send lock: so_rcv -> radix node head All protocol locks before socket locks: unp -> so_snd udp -> udpinp -> so_snd tcp -> tcpinp -> so_snd
2004-06-13 00:23:03 +00:00
{ "so_rcv", &lock_class_mtx_sleep },
- convert radix node head lock from mutex to rwlock - make radix node head lock not recursive - fix LOR in rtexpunge - fix LOR in rtredirect Reviewed by: sam
2008-12-07 21:15:43 +00:00
{ "radix node head", &lock_class_rw },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
{ "rtentry", &lock_class_mtx_sleep },
{ "ifaddr", &lock_class_mtx_sleep },
{ NULL, NULL },
Introduce in_multi_mtx, which will protect IPv4-layer multicast address lists, as well as accessor macros. For now, this is a recursive mutex due code sequences where IPv4 multicast calls into IGMP calls into ip_output(), which then tests for a multicast forwarding case. For support macros in in_var.h to check multicast address lists, assert that in_multi_mtx is held. Acquire in_multi_mtx around iteration over the IPv4 multicast address lists, such as in ip_input() and ip_output(). Acquire in_multi_mtx when manipulating the IPv4 layer multicast addresses, as well as over the manipulation of ifnet multicast address lists in order to keep the two layers in sync. Lock down accesses to IPv4 multicast addresses in IGMP, or assert the lock when performing IGMP join/leave events. Eliminate spl's associated with IPv4 multicast addresses, portions of IGMP that weren't previously expunged by IGMP locking. Add in_multi_mtx, igmp_mtx, and if_addr_mtx lock order to hard-coded lock order in WITNESS, in that order. Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca> MFC after: 10 days
2005-08-03 19:29:47 +00:00
/*
Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: import from p4 bms_netdev. Summary of changes: * Connect netinet6/in6_mcast.c to build. The legacy KAME KPIs are mostly preserved. * Eliminate now dead code from ip6_output.c. Don't do mbuf bingo, we are not going to do RFC 2292 style CMSG tricks for multicast options as they are not required by any current IPv6 normative reference. * Refactor transports (UDP, raw_ip6) to do own mcast filtering. SCTP, TCP unaffected by this change. * Add ip6_msource, in6_msource structs to in6_var.h. * Hookup mld_ifinfo state to in6_ifextra, allocate from domifattach path. * Eliminate IN6_LOOKUP_MULTI(), it is no longer referenced. Kernel consumers which need this should use in6m_lookup(). * Refactor IPv6 socket group memberships to use a vector (like IPv4). * Update ifmcstat(8) for IPv6 SSM. * Add witness lock order for IN6_MULTI_LOCK. * Move IN6_MULTI_LOCK out of lower ip6_output()/ip6_input() paths. * Introduce IP6STAT_ADD/SUB/INC/DEC as per rwatson's IPv4 cleanup. * Update carp(4) for new IPv6 SSM KPIs. * Virtualize ip6_mrouter socket. Changes mostly localized to IPv6 MROUTING. * Don't do a local group lookup in MROUTING. * Kill unused KAME prototypes in6_purgemkludge(), in6_restoremkludge(). * Preserve KAME DAD timer jitter behaviour in MLDv1 compatibility mode. * Bump __FreeBSD_version to 800084. * Update UPDATING. NOTE WELL: * This code hasn't been tested against real MLDv2 queriers (yet), although the on-wire protocol has been verified in Wireshark. * There are a few unresolved issues in the socket layer APIs to do with scope ID propagation. * There is a LOR present in ip6_output()'s use of in6_setscope() which needs to be resolved. See comments in mld6.c. This is believed to be benign and can't be avoided for the moment without re-introducing an indirect netisr. This work was mostly derived from the IGMPv3 implementation, and has been sponsored by a third party.
2009-04-29 19:19:13 +00:00
* IPv4 multicast:
* protocol locks before interface locks, after UDP locks.
Introduce in_multi_mtx, which will protect IPv4-layer multicast address lists, as well as accessor macros. For now, this is a recursive mutex due code sequences where IPv4 multicast calls into IGMP calls into ip_output(), which then tests for a multicast forwarding case. For support macros in in_var.h to check multicast address lists, assert that in_multi_mtx is held. Acquire in_multi_mtx around iteration over the IPv4 multicast address lists, such as in ip_input() and ip_output(). Acquire in_multi_mtx when manipulating the IPv4 layer multicast addresses, as well as over the manipulation of ifnet multicast address lists in order to keep the two layers in sync. Lock down accesses to IPv4 multicast addresses in IGMP, or assert the lock when performing IGMP join/leave events. Eliminate spl's associated with IPv4 multicast addresses, portions of IGMP that weren't previously expunged by IGMP locking. Add in_multi_mtx, igmp_mtx, and if_addr_mtx lock order to hard-coded lock order in WITNESS, in that order. Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca> MFC after: 10 days
2005-08-03 19:29:47 +00:00
*/
Convert pcbinfo and inpcb mutexes to rwlocks, and modify macros to explicitly select write locking for all use of the inpcb mutex. Update some pcbinfo lock assertions to assert locked rather than write-locked, although in practice almost all uses of the pcbinfo rwlock main exclusive, and all instances of inpcb lock acquisition are exclusive. This change should introduce (ideally) little functional change. However, it lays the groundwork for significantly increased parallelism in the TCP/IP code. MFC after: 3 months Tested by: kris (superset of committered patch)
2008-04-17 21:38:18 +00:00
{ "udpinp", &lock_class_rw },
Introduce in_multi_mtx, which will protect IPv4-layer multicast address lists, as well as accessor macros. For now, this is a recursive mutex due code sequences where IPv4 multicast calls into IGMP calls into ip_output(), which then tests for a multicast forwarding case. For support macros in in_var.h to check multicast address lists, assert that in_multi_mtx is held. Acquire in_multi_mtx around iteration over the IPv4 multicast address lists, such as in ip_input() and ip_output(). Acquire in_multi_mtx when manipulating the IPv4 layer multicast addresses, as well as over the manipulation of ifnet multicast address lists in order to keep the two layers in sync. Lock down accesses to IPv4 multicast addresses in IGMP, or assert the lock when performing IGMP join/leave events. Eliminate spl's associated with IPv4 multicast addresses, portions of IGMP that weren't previously expunged by IGMP locking. Add in_multi_mtx, igmp_mtx, and if_addr_mtx lock order to hard-coded lock order in WITNESS, in that order. Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca> MFC after: 10 days
2005-08-03 19:29:47 +00:00
{ "in_multi_mtx", &lock_class_mtx_sleep },
{ "igmp_mtx", &lock_class_mtx_sleep },
Convert the per-interface address list lock from a mutex to a reader/writer lock. Reviewed by: bz
2012-01-09 19:34:12 +00:00
{ "if_addr_lock", &lock_class_rw },
Introduce in_multi_mtx, which will protect IPv4-layer multicast address lists, as well as accessor macros. For now, this is a recursive mutex due code sequences where IPv4 multicast calls into IGMP calls into ip_output(), which then tests for a multicast forwarding case. For support macros in in_var.h to check multicast address lists, assert that in_multi_mtx is held. Acquire in_multi_mtx around iteration over the IPv4 multicast address lists, such as in ip_input() and ip_output(). Acquire in_multi_mtx when manipulating the IPv4 layer multicast addresses, as well as over the manipulation of ifnet multicast address lists in order to keep the two layers in sync. Lock down accesses to IPv4 multicast addresses in IGMP, or assert the lock when performing IGMP join/leave events. Eliminate spl's associated with IPv4 multicast addresses, portions of IGMP that weren't previously expunged by IGMP locking. Add in_multi_mtx, igmp_mtx, and if_addr_mtx lock order to hard-coded lock order in WITNESS, in that order. Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca> MFC after: 10 days
2005-08-03 19:29:47 +00:00
{ NULL, NULL },
Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: import from p4 bms_netdev. Summary of changes: * Connect netinet6/in6_mcast.c to build. The legacy KAME KPIs are mostly preserved. * Eliminate now dead code from ip6_output.c. Don't do mbuf bingo, we are not going to do RFC 2292 style CMSG tricks for multicast options as they are not required by any current IPv6 normative reference. * Refactor transports (UDP, raw_ip6) to do own mcast filtering. SCTP, TCP unaffected by this change. * Add ip6_msource, in6_msource structs to in6_var.h. * Hookup mld_ifinfo state to in6_ifextra, allocate from domifattach path. * Eliminate IN6_LOOKUP_MULTI(), it is no longer referenced. Kernel consumers which need this should use in6m_lookup(). * Refactor IPv6 socket group memberships to use a vector (like IPv4). * Update ifmcstat(8) for IPv6 SSM. * Add witness lock order for IN6_MULTI_LOCK. * Move IN6_MULTI_LOCK out of lower ip6_output()/ip6_input() paths. * Introduce IP6STAT_ADD/SUB/INC/DEC as per rwatson's IPv4 cleanup. * Update carp(4) for new IPv6 SSM KPIs. * Virtualize ip6_mrouter socket. Changes mostly localized to IPv6 MROUTING. * Don't do a local group lookup in MROUTING. * Kill unused KAME prototypes in6_purgemkludge(), in6_restoremkludge(). * Preserve KAME DAD timer jitter behaviour in MLDv1 compatibility mode. * Bump __FreeBSD_version to 800084. * Update UPDATING. NOTE WELL: * This code hasn't been tested against real MLDv2 queriers (yet), although the on-wire protocol has been verified in Wireshark. * There are a few unresolved issues in the socket layer APIs to do with scope ID propagation. * There is a LOR present in ip6_output()'s use of in6_setscope() which needs to be resolved. See comments in mld6.c. This is believed to be benign and can't be avoided for the moment without re-introducing an indirect netisr. This work was mostly derived from the IGMPv3 implementation, and has been sponsored by a third party.
2009-04-29 19:19:13 +00:00
/*
* IPv6 multicast:
* protocol locks before interface locks, after UDP locks.
*/
{ "udpinp", &lock_class_rw },
{ "in6_multi_mtx", &lock_class_mtx_sleep },
{ "mld_mtx", &lock_class_mtx_sleep },
Convert the per-interface address list lock from a mutex to a reader/writer lock. Reviewed by: bz
2012-01-09 19:34:12 +00:00
{ "if_addr_lock", &lock_class_rw },
Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: import from p4 bms_netdev. Summary of changes: * Connect netinet6/in6_mcast.c to build. The legacy KAME KPIs are mostly preserved. * Eliminate now dead code from ip6_output.c. Don't do mbuf bingo, we are not going to do RFC 2292 style CMSG tricks for multicast options as they are not required by any current IPv6 normative reference. * Refactor transports (UDP, raw_ip6) to do own mcast filtering. SCTP, TCP unaffected by this change. * Add ip6_msource, in6_msource structs to in6_var.h. * Hookup mld_ifinfo state to in6_ifextra, allocate from domifattach path. * Eliminate IN6_LOOKUP_MULTI(), it is no longer referenced. Kernel consumers which need this should use in6m_lookup(). * Refactor IPv6 socket group memberships to use a vector (like IPv4). * Update ifmcstat(8) for IPv6 SSM. * Add witness lock order for IN6_MULTI_LOCK. * Move IN6_MULTI_LOCK out of lower ip6_output()/ip6_input() paths. * Introduce IP6STAT_ADD/SUB/INC/DEC as per rwatson's IPv4 cleanup. * Update carp(4) for new IPv6 SSM KPIs. * Virtualize ip6_mrouter socket. Changes mostly localized to IPv6 MROUTING. * Don't do a local group lookup in MROUTING. * Kill unused KAME prototypes in6_purgemkludge(), in6_restoremkludge(). * Preserve KAME DAD timer jitter behaviour in MLDv1 compatibility mode. * Bump __FreeBSD_version to 800084. * Update UPDATING. NOTE WELL: * This code hasn't been tested against real MLDv2 queriers (yet), although the on-wire protocol has been verified in Wireshark. * There are a few unresolved issues in the socket layer APIs to do with scope ID propagation. * There is a LOR present in ip6_output()'s use of in6_setscope() which needs to be resolved. See comments in mld6.c. This is believed to be benign and can't be avoided for the moment without re-introducing an indirect netisr. This work was mostly derived from the IGMPv3 implementation, and has been sponsored by a third party.
2009-04-29 19:19:13 +00:00
{ NULL, NULL },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
/*
* UNIX Domain Sockets
- Comment out NULL, NULL barrier for Unix domain sockets section as the double NULL entries signal Witness to stop processing the array of order entries meaning none of the spin locks are added resulting in panics on boot. - Add a missing NULL, NULL terminator to the Slip locks list to keep them separate from the spin locks.
2004-06-03 20:07:44 +00:00
*/
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
{ "unp_global_rwlock", &lock_class_rw },
{ "unp_list_lock", &lock_class_mtx_sleep },
Introduce socket and UNIX domain socket locks into hard-coded lock order definition for witness. Send lock before receive lock, and socket locks after accept but before select: filedesc -> accept -> so_snd -> so_rcv -> sellck All routing locks after send lock: so_rcv -> radix node head All protocol locks before socket locks: unp -> so_snd udp -> udpinp -> so_snd tcp -> tcpinp -> so_snd
2004-06-13 00:23:03 +00:00
{ "unp", &lock_class_mtx_sleep },
{ "so_snd", &lock_class_mtx_sleep },
{ NULL, NULL },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
/*
* UDP/IP
*/
Convert pcbinfo and inpcb mutexes to rwlocks, and modify macros to explicitly select write locking for all use of the inpcb mutex. Update some pcbinfo lock assertions to assert locked rather than write-locked, although in practice almost all uses of the pcbinfo rwlock main exclusive, and all instances of inpcb lock acquisition are exclusive. This change should introduce (ideally) little functional change. However, it lays the groundwork for significantly increased parallelism in the TCP/IP code. MFC after: 3 months Tested by: kris (superset of committered patch)
2008-04-17 21:38:18 +00:00
{ "udp", &lock_class_rw },
{ "udpinp", &lock_class_rw },
Introduce socket and UNIX domain socket locks into hard-coded lock order definition for witness. Send lock before receive lock, and socket locks after accept but before select: filedesc -> accept -> so_snd -> so_rcv -> sellck All routing locks after send lock: so_rcv -> radix node head All protocol locks before socket locks: unp -> so_snd udp -> udpinp -> so_snd tcp -> tcpinp -> so_snd
2004-06-13 00:23:03 +00:00
{ "so_snd", &lock_class_mtx_sleep },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
{ NULL, NULL },
/*
* TCP/IP
*/
Convert pcbinfo and inpcb mutexes to rwlocks, and modify macros to explicitly select write locking for all use of the inpcb mutex. Update some pcbinfo lock assertions to assert locked rather than write-locked, although in practice almost all uses of the pcbinfo rwlock main exclusive, and all instances of inpcb lock acquisition are exclusive. This change should introduce (ideally) little functional change. However, it lays the groundwork for significantly increased parallelism in the TCP/IP code. MFC after: 3 months Tested by: kris (superset of committered patch)
2008-04-17 21:38:18 +00:00
{ "tcp", &lock_class_rw },
{ "tcpinp", &lock_class_rw },
Introduce socket and UNIX domain socket locks into hard-coded lock order definition for witness. Send lock before receive lock, and socket locks after accept but before select: filedesc -> accept -> so_snd -> so_rcv -> sellck All routing locks after send lock: so_rcv -> radix node head All protocol locks before socket locks: unp -> so_snd udp -> udpinp -> so_snd tcp -> tcpinp -> so_snd
2004-06-13 00:23:03 +00:00
{ "so_snd", &lock_class_mtx_sleep },
Expand the hard-coded WITNESS lock order to include the following relationships: Sockets: filedesc->accept->sellck Routing: radix node head->rtentry->ifaddr UDP: udp->udpinp TCP: tcp->tcpinp SLIP: slip_mtx->slip sc_mtx Drop in a place holder section for UNIX domain sockets. Various sections to be expanded over the next few days.
2004-06-02 23:28:06 +00:00
{ NULL, NULL },
Add netatalk mutexes to hard-coded WITNESS lock order.
2004-07-25 20:16:51 +00:00
/*
* netatalk
*/
{ "ddp_list_mtx", &lock_class_mtx_sleep },
{ "ddp_mtx", &lock_class_mtx_sleep },
{ NULL, NULL },
Hard code witness lock order for BPF locks.
2004-09-09 05:01:37 +00:00
/*
* BPF
*/
{ "bpf global lock", &lock_class_mtx_sleep },
- Improve BPF locking model. Interface locks and descriptor locks are converted from mutex(9) to rwlock(9). This greately improves performance: in most common case we need to acquire 1 reader lock instead of 2 mutexes. - Remove filter(descriptor) (reader) lock in bpf_mtap[2] This was suggested by glebius@. We protect filter by requesting interface writer lock on filter change. - Cover struct bpf_if under BPF_INTERNAL define. This permits including bpf.h without including rwlock stuff. However, this is is temporary solution, struct bpf_if should be made opaque for any external caller. Found by: Dmitrij Tejblum <tejblum@yandex-team.ru> Sponsored by: Yandex LLC Reviewed by: glebius (previous version) Reviewed by: silence on -net@ Approved by: (mentor) MFC after: 3 weeks
2012-04-06 06:53:58 +00:00
{ "bpf interface lock", &lock_class_rw },
Fix old panic when BPF consumer attaches to destroying interface. 'flags' field is added to the end of bpf_if structure. Currently the only flag is BPFIF_FLAG_DYING which is set on bpf detach and checked by bpf_attachd() Problem can be easily triggered on SMP stable/[89] by the following command (sort of): 'while true; do ifconfig vlan222 create vlan 222 vlandev em0 up ; tcpdump -pi vlan222 & ; ifconfig vlan222 destroy ; done' Fix possible use-after-free when BPF detaches itself from interface, freeing bpf_bif memory, while interface is still UP and there can be routes via this interface. Freeing is now delayed till ifnet_departure_event is received via eventhandler(9) api. Convert bpfd rwlock back to mutex due lack of performance gain (currently checking if packet matches filter is done without holding bpfd lock and we have to acquire write lock if packet matches) Approved by: kib(mentor) MFC in: 4 weeks
2012-05-21 22:17:29 +00:00
{ "bpf cdev lock", &lock_class_mtx_sleep },
Add a missing terminator. Confirmed by: rwatson
2005-04-09 11:31:31 +00:00
{ NULL, NULL },
Document, via WITNESS, that the NFS server mutex falls ahead of the socket buffer mutexes.
2005-03-09 21:38:53 +00:00
/*
* NFS server
*/
{ "nfsd_mtx", &lock_class_mtx_sleep },
{ "so_snd", &lock_class_mtx_sleep },
Hard code witness lock order for BPF locks.
2004-09-09 05:01:37 +00:00
{ NULL, NULL },
Update 802.11 wireless support: o major overhaul of the way channels are handled: channels are now fully enumerated and uniquely identify the operating characteristics; these changes are visible to user applications which require changes o make scanning support independent of the state machine to enable background scanning and roaming o move scanning support into loadable modules based on the operating mode to enable different policies and reduce the memory footprint on systems w/ constrained resources o add background scanning in station mode (no support for adhoc/ibss mode yet) o significantly speedup sta mode scanning with a variety of techniques o add roaming support when background scanning is supported; for now we use a simple algorithm to trigger a roam: we threshold the rssi and tx rate, if either drops too low we try to roam to a new ap o add tx fragmentation support o add first cut at 802.11n support: this code works with forthcoming drivers but is incomplete; it's included now to establish a baseline for other drivers to be developed and for user applications o adjust max_linkhdr et. al. to reflect 802.11 requirements; this eliminates prepending mbufs for traffic generated locally o add support for Atheros protocol extensions; mainly the fast frames encapsulation (note this can be used with any card that can tx+rx large frames correctly) o add sta support for ap's that beacon both WPA1+2 support o change all data types from bsd-style to posix-style o propagate noise floor data from drivers to net80211 and on to user apps o correct various issues in the sta mode state machine related to handling authentication and association failures o enable the addition of sta mode power save support for drivers that need net80211 support (not in this commit) o remove old WI compatibility ioctls (wicontrol is officially dead) o change the data structures returned for get sta info and get scan results so future additions will not break user apps o fixed tx rate is now maintained internally as an ieee rate and not an index into the rate set; this needs to be extended to deal with multi-mode operation o add extended channel specifications to radiotap to enable 11n sniffing Drivers: o ath: add support for bg scanning, tx fragmentation, fast frames, dynamic turbo (lightly tested), 11n (sniffing only and needs new hal) o awi: compile tested only o ndis: lightly tested o ipw: lightly tested o iwi: add support for bg scanning (well tested but may have some rough edges) o ral, ural, rum: add suppoort for bg scanning, calibrate rssi data o wi: lightly tested This work is based on contributions by Atheros, kmacy, sephe, thompsa, mlaier, kevlo, and others. Much of the scanning work was supported by Atheros. The 11n work was supported by Marvell.
2007-06-11 03:36:55 +00:00
/*
* IEEE 802.11
*/
{ "802.11 com lock", &lock_class_mtx_sleep},
{ NULL, NULL },
/*
* Network drivers
*/
{ "network driver", &lock_class_mtx_sleep},
{ NULL, NULL },
ng_node and ng_worklist locks both migrated from being spinning locks to adaptive mutexes. Let witness(4) calm down and bring proper types of those locks to the lock order database. Glanced at by: rwatson
2007-04-01 15:48:10 +00:00
/*
* Netgraph
*/
{ "ng_node", &lock_class_mtx_sleep },
{ "ng_worklist", &lock_class_mtx_sleep },
{ NULL, NULL },
CDEV lock should be before 'system map' lock. Hardcode this order to help track down reported LOR. LOR reported by: Thierry Herbelot <thierry@herbelot.com> LOR info: http://sources.zabbadoz.net/freebsd/lor.html#080
2005-04-09 13:32:01 +00:00
/*
* CDEV
*/
{ "system map", &lock_class_mtx_sleep },
- Define the real lock order with cdev and a few vm/vfs related locks. This can be removed once cdev no longer calls free() with the cdev lock held.
2005-04-22 22:43:31 +00:00
{ "vm page queue mutex", &lock_class_mtx_sleep },
{ "vnode interlock", &lock_class_mtx_sleep },
{ "cdev", &lock_class_mtx_sleep },
CDEV lock should be before 'system map' lock. Hardcode this order to help track down reported LOR. LOR reported by: Thierry Herbelot <thierry@herbelot.com> LOR info: http://sources.zabbadoz.net/freebsd/lor.html#080
2005-04-09 13:32:01 +00:00
{ NULL, NULL },
On Alan's advice, rather than do a wholesale conversion on a single architecture from page queue lock to a hashed array of page locks (based on a patch by Jeff Roberson), I've implemented page lock support in the MI code and have only moved vm_page's hold_count out from under page queue mutex to page lock. This changes pmap_extract_and_hold on all pmaps. Supported by: Bitgravity Inc. Discussed with: alc, jeffr, and kib
2010-04-30 00:46:43 +00:00
/*
* VM
*
*/
{ "vm object", &lock_class_mtx_sleep },
{ "page lock", &lock_class_mtx_sleep },
{ "vm page queue mutex", &lock_class_mtx_sleep },
{ "pmap", &lock_class_mtx_sleep },
{ NULL, NULL },
Record kqueue -> struct mount mtx -> vnode interlock lock order to catch the places where reverse lock order is instantiated. OKed by: jeff
2007-02-02 09:02:18 +00:00
/*
* kqueue/VFS interaction
*/
{ "kqueue", &lock_class_mtx_sleep },
{ "struct mount mtx", &lock_class_mtx_sleep },
{ "vnode interlock", &lock_class_mtx_sleep },
{ NULL, NULL },
Add minimal ZFS lock hierarchy
2009-05-20 02:51:48 +00:00
/*
* ZFS locking
*/
{ "dn->dn_mtx", &lock_class_sx },
{ "dr->dt.di.dr_mtx", &lock_class_sx },
{ "db->db_mtx", &lock_class_sx },
{ NULL, NULL },
- Add a comment at the start of the spin locks list. - The alpha SMP code uses an "ap boot" spinlock as well.
2001-04-13 08:31:38 +00:00
/*
* spin locks
*/
Move the ap boot spin lock earlier in the lock order before the sio(4) lock since we occasionally call printf() while holding the ap boot lock which can call down into the sio(4) driver if using a serial console.
2001-10-01 22:50:30 +00:00
#ifdef SMP
{ "ap boot", &lock_class_mtx_spin },
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
#endif
Add the spin lock used by the binary nvidia driver to the static lock order list so that WITNESS and the driver play together nicely. Tested by: Harald Schmalzbauer MFC after: 3 days
2005-09-26 18:30:12 +00:00
{ "rm.mutex_mtx", &lock_class_mtx_spin },
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ "sio", &lock_class_mtx_spin },
Fix the spinlock static table adding missing spinlocks. - rm_spinlock has turnstile chain as child - srclock has callout and clk as child, found by witness "emulation". Just move it very high in our ranking
2007-11-24 04:32:32 +00:00
{ "scrlock", &lock_class_mtx_spin },
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
#ifdef __i386__
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ "cy", &lock_class_mtx_spin },
- Remove zstty spin lock for no longer existing zs(4). - Move the rtc_mtx spin lock out from under #ifdef SMP as it's just not SMP-specific. - Add a new spin lock pcib_mtx for locking "fast" interrupt handlers of host-to-PCI bridge drivers on sparc64.
2007-06-16 23:30:57 +00:00
#endif
#ifdef __sparc64__
{ "pcib_mtx", &lock_class_mtx_spin },
{ "rtc_mtx", &lock_class_mtx_spin },
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
#endif
Add the scc_hwmtx spin mutex, defined by scc(4).
2006-04-07 22:15:54 +00:00
{ "scc_hwmtx", &lock_class_mtx_spin },
Register the uart(4)'s spin lock with witness(4).
2004-01-25 15:04:37 +00:00
{ "uart_hwmtx", &lock_class_mtx_spin },
The "taskqueue_fast" spinlocks were renamed to "fast_taskqueue" in subr_taskqueue.c:r1.32 Reported by: rdivacky
2006-08-26 11:21:25 +00:00
{ "fast_taskqueue", &lock_class_mtx_spin },
Update spin lock order list for new i386 interrupt and SMP code.
2003-11-03 22:38:30 +00:00
{ "intr table", &lock_class_mtx_spin },
Fix witness(4) warnings about mutex use. Group mutexes used in hwpmc(4) into 3 "types" in the sense of witness(4): - leaf spin mutexes---only one of these should be held at a time, so these mutexes are specified as belonging to a single witness type "pmc-leaf". - `struct pmc_owner' descriptors are protected by a spin mutex of witness type "pmc-owner-proc". Since we call wakeup_one() while holding these mutexes, the witness type of these mutexes needs to dominate that of "sleepq chain" mutexes. - logger threads use a sleep mutex, of type "pmc-sleep". Submitted by: wkoszek (earlier patch)
2007-04-19 08:02:51 +00:00
#ifdef HWPMC_HOOKS
{ "pmc-per-proc", &lock_class_mtx_spin },
#endif
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
{ "process slock", &lock_class_mtx_spin },
Add an implementation of a generic sleep queue abstraction that is used to queue threads sleeping on a wait channel similar to how turnstiles are used to queue threads waiting for a lock. This subsystem will be used as the backend for sleep/wakeup and condition variables initially. Eventually it will also be used to replace the ithread-specific iwait thread inhibitor. Sleep queues are also not locked by sched_lock, so this splits sched_lock up a bit further increasing concurrency within the scheduler. Sleep queues also natively support timeouts on sleeps and interruptible sleeps allowing for the reduction of a lot of duplicated code between the sleep/wakeup and condition variable implementations. For more details on the sleep queue implementation, check the comments in sys/sleepqueue.h and kern/subr_sleepqueue.c.
2004-02-27 18:33:09 +00:00
{ "sleepq chain", &lock_class_mtx_spin },
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
{ "umtx lock", &lock_class_mtx_spin },
Fix the spinlock static table adding missing spinlocks. - rm_spinlock has turnstile chain as child - srclock has callout and clk as child, found by witness "emulation". Just move it very high in our ranking
2007-11-24 04:32:32 +00:00
{ "rm_spinlock", &lock_class_mtx_spin },
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
{ "turnstile chain", &lock_class_mtx_spin },
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
{ "turnstile lock", &lock_class_mtx_spin },
{ "sched lock", &lock_class_mtx_spin },
Add an implementation of turnstiles and change the sleep mutex code to use turnstiles to implement blocking isntead of implementing a thread queue directly. These turnstiles are somewhat similar to those used in Solaris 7 as described in Solaris Internals but are also different. Turnstiles do not come out of a fixed-sized pool. Rather, each thread is assigned a turnstile when it is created that it frees when it is destroyed. When a thread blocks on a lock, it donates its turnstile to that lock to serve as queue of blocked threads. The queue associated with a given lock is found by a lookup in a simple hash table. The turnstile itself is protected by a lock associated with its entry in the hash table. This means that sched_lock is no longer needed to contest on a mutex. Instead, sched_lock is only used when manipulating run queues or thread priorities. Turnstiles also implement priority propagation inherently. Currently turnstiles only support mutexes. Eventually, however, turnstiles may grow two queue's to support a non-sleepable reader/writer lock implementation. For more details, see the comments in sys/turnstile.h and kern/subr_turnstile.c. The two primary advantages from the turnstile code include: 1) the size of struct mutex shrinks by four pointers as it no longer stores the thread queue linkages directly, and 2) less contention on sched_lock in SMP systems including the ability for multiple CPUs to contend on different locks simultaneously (not that this last detail is necessarily that much of a big win). Note that 1) means that this commit is a kernel ABI breaker, so don't mix old modules with a new kernel and vice versa. Tested on: i386 SMP, sparc64 SMP, alpha SMP
2003-11-11 22:07:29 +00:00
{ "td_contested", &lock_class_mtx_spin },
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ "callout", &lock_class_mtx_spin },
Add entropy harvest mutex to hard-coded spin lock witness lock order, remove previous entropy harvesting mutex names as they are no longer present. Commit to this file was ommitted when randomdev_soft.c:1.5 was made. Feet shot: Robert Huff <roberthuff at rcn dot com>
2004-10-11 08:26:18 +00:00
{ "entropy harvest mutex", &lock_class_mtx_spin },
Introduce a spinlock for synchronizing access to the video output hardware in syscons. This replaces a simple access semaphore that was assumed to be protected by Giant but often was not. If two threads that were otherwise SMP-safe called printf at the same time, there was a high likelyhood that the semaphore would get corrupted and result in a permanently frozen video console. This is similar to what is already done in the serial console drivers.
2006-09-13 15:48:15 +00:00
{ "syscons video lock", &lock_class_mtx_spin },
Fix some entries in the locks static table of witness. In particular: - smp_tlb_mtx is no longer used, so it is axed. - smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in the table, however, has been the source of a false positive LOR reporting with the dt_lock. However, smp rendezvous lock would have had sched_lock there for older lock, so it wasn't still a leaf lock. - allpmaps is only used in ia32 architecture, so it is inserted in the appropriate stub. Addictionally: - kse_zombie_lock is no longer present, so its definition is axed out. - zombie_lock doesn't need to have an exported symbol, so just let's it be declared as static. Tested by: kris Approved by: jeff (mentor) Approved by: re
2007-09-20 20:38:43 +00:00
#ifdef SMP
{ "smp rendezvous", &lock_class_mtx_spin },
Initial support for Freescale PowerQUICC III MPC85xx system-on-chip family. The PQ3 is a high performance integrated communications processing system based on the e500 core, which is an embedded RISC processor that implements the 32-bit Book E definition of the PowerPC architecture. For details refer to: http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=MPC8555E This port was tested and successfully run on the following members of the PQ3 family: MPC8533, MPC8541, MPC8548, MPC8555. The following major integrated peripherals are supported: * On-chip peripherals bus * OpenPIC interrupt controller * UART * Ethernet (TSEC) * Host/PCI bridge * QUICC engine (SCC functionality) This commit brings the main functionality and will be followed by individual drivers that are logically separate from this base. Approved by: cognet (mentor) Obtained from: Juniper, Semihalf MFp4: e500
2008-03-03 17:17:00 +00:00
#endif
#ifdef __powerpc__
{ "tlb0", &lock_class_mtx_spin },
Fix some entries in the locks static table of witness. In particular: - smp_tlb_mtx is no longer used, so it is axed. - smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in the table, however, has been the source of a false positive LOR reporting with the dt_lock. However, smp rendezvous lock would have had sched_lock there for older lock, so it wasn't still a leaf lock. - allpmaps is only used in ia32 architecture, so it is inserted in the appropriate stub. Addictionally: - kse_zombie_lock is no longer present, so its definition is axed out. - zombie_lock doesn't need to have an exported symbol, so just let's it be declared as static. Tested by: kris Approved by: jeff (mentor) Approved by: re
2007-09-20 20:38:43 +00:00
#endif
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*
* leaf locks
*/
Add missing stubs for spinlocks cpuset and intrcnt. Submitted by: kris
2008-04-12 13:51:18 +00:00
{ "intrcnt", &lock_class_mtx_spin },
Introduce a standard name for the lock protecting an interrupt controller and it's associated state variables: icu_lock with the name "icu". This renames the imen_mtx for x86 SMP, but also uses the lock to protect access to the 8259 PIC on x86 UP. This also adds an appropriate lock to the various Alpha chipsets which fixes problems with Alpha SMP machines dropping interrupts with an SMP kernel.
2001-12-20 23:48:31 +00:00
{ "icu", &lock_class_mtx_spin },
Fix some entries in the locks static table of witness. In particular: - smp_tlb_mtx is no longer used, so it is axed. - smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in the table, however, has been the source of a false positive LOR reporting with the dt_lock. However, smp rendezvous lock would have had sched_lock there for older lock, so it wasn't still a leaf lock. - allpmaps is only used in ia32 architecture, so it is inserted in the appropriate stub. Addictionally: - kse_zombie_lock is no longer present, so its definition is axed out. - zombie_lock doesn't need to have an exported symbol, so just let's it be declared as static. Tested by: kris Approved by: jeff (mentor) Approved by: re
2007-09-20 20:38:43 +00:00
#if defined(SMP) && defined(__sparc64__)
- Add a spin lock to single thread cache invalidation and tlb flush ipis, which allows ipis to be sent outside of Giant. - Remove the ap boot mutex, which is unused.
2002-12-22 20:50:23 +00:00
{ "ipi", &lock_class_mtx_spin },
#endif
Fix some entries in the locks static table of witness. In particular: - smp_tlb_mtx is no longer used, so it is axed. - smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in the table, however, has been the source of a false positive LOR reporting with the dt_lock. However, smp rendezvous lock would have had sched_lock there for older lock, so it wasn't still a leaf lock. - allpmaps is only used in ia32 architecture, so it is inserted in the appropriate stub. Addictionally: - kse_zombie_lock is no longer present, so its definition is axed out. - zombie_lock doesn't need to have an exported symbol, so just let's it be declared as static. Tested by: kris Approved by: jeff (mentor) Approved by: re
2007-09-20 20:38:43 +00:00
#ifdef __i386__
{ "allpmaps", &lock_class_mtx_spin },
{ "descriptor tables", &lock_class_mtx_spin },
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
#endif
- Move the 'clk' spinlock below other spin locks since KTR trace events may need the clock lock for nanotime(). - Add KTR trace events for lock list manipulations and other witness operations. - Use a temporary variable instead of setting the lock list head directly and then setting up the links to add a new lock list entry to the lock list. This small race could result in witness "forgetting" about all the locks held by this process temporarily during an interrupt. - Close a more fatal race condition when removing a lock from a list. Removing a lock from the list entails both decrementing the count of items in this bucket as well as shuffling items in the current bucket up a notch to replace the gap left by the removed item. Wrap these operations in a critical section.
2001-06-25 23:17:52 +00:00
{ "clk", &lock_class_mtx_spin },
Add missing stubs for spinlocks cpuset and intrcnt. Submitted by: kris
2008-04-12 13:51:18 +00:00
{ "cpuset", &lock_class_mtx_spin },
Fix some entries in the locks static table of witness. In particular: - smp_tlb_mtx is no longer used, so it is axed. - smp rendezvous lock isn't really a leaf spin-mutex. Its bad placement in the table, however, has been the source of a false positive LOR reporting with the dt_lock. However, smp rendezvous lock would have had sched_lock there for older lock, so it wasn't still a leaf lock. - allpmaps is only used in ia32 architecture, so it is inserted in the appropriate stub. Addictionally: - kse_zombie_lock is no longer present, so its definition is axed out. - zombie_lock doesn't need to have an exported symbol, so just let's it be declared as static. Tested by: kris Approved by: jeff (mentor) Approved by: re
2007-09-20 20:38:43 +00:00
{ "mprof lock", &lock_class_mtx_spin },
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
{ "zombie lock", &lock_class_mtx_spin },
- Tell witness about ALQ's spin lock.
2002-09-22 07:11:57 +00:00
{ "ALD Queue", &lock_class_mtx_spin },
Register the machine check private state spinlock on ia64.
2002-10-12 00:33:36 +00:00
#ifdef __ia64__
{ "MCA spin lock", &lock_class_mtx_spin },
Initiate de-orbit burn for USE_PCI_BIOS_FOR_READ_WRITE. This has been #if'ed out for a while. Complete the deed and tidy up some other bits. We need to be able to call this stuff from outer edges of interrupt handlers for devices that have the ISR bits in pci config space. Making the bios code mpsafe was just too hairy. We had also stubbed it out some time ago due to there simply being too much brokenness in too many systems. This adds a leaf lock so that it is safe to use pci_read_config() and pci_write_config() from interrupt handlers. We still will use pcibios to do interrupt routing if there is no acpi.. [yes, I tested this] Briefly glanced at by: imp
2003-02-18 03:36:49 +00:00
#endif
Add __amd64__ to the ifdefs that introduce the "pcicfg" spinlock to witness. Approved by: re (safe amd64 support)
2003-05-31 06:42:37 +00:00
#if defined(__i386__) || defined(__amd64__)
Initiate de-orbit burn for USE_PCI_BIOS_FOR_READ_WRITE. This has been #if'ed out for a while. Complete the deed and tidy up some other bits. We need to be able to call this stuff from outer edges of interrupt handlers for devices that have the ISR bits in pci config space. Making the bios code mpsafe was just too hairy. We had also stubbed it out some time ago due to there simply being too much brokenness in too many systems. This adds a leaf lock so that it is safe to use pci_read_config() and pci_write_config() from interrupt handlers. We still will use pcibios to do interrupt routing if there is no acpi.. [yes, I tested this] Briefly glanced at by: imp
2003-02-18 03:36:49 +00:00
{ "pcicfg", &lock_class_mtx_spin },
When you call MiniportInitialize() for an 802.11 driver, it will at some point result in a status event being triggered (it should be a link down event: the Microsoft driver design guide says you should generate one when the NIC is initialized). Some drivers generate the event during MiniportInitialize(), such that by the time MiniportInitialize() completes, the NIC is ready to go. But some drivers, in particular the ones for Atheros wireless NICs, don't generate the event until after a device interrupt occurs at some point after MiniportInitialize() has completed. The gotcha is that you have to wait until the link status event occurs one way or the other before you try to fiddle with any settings (ssid, channel, etc...). For the drivers that set the event sycnhronously this isn't a problem, but for the others we have to pause after calling ndis_init_nic() and wait for the event to arrive before continuing. Failing to wait can cause big trouble: on my SMP system, calling ndis_setstate_80211() after ndis_init_nic() completes, but _before_ the link event arrives, will lock up or reset the system. What we do now is check to see if a link event arrived while ndis_init_nic() was running, and if it didn't we msleep() until it does. Along the way, I discovered a few other problems: - Defered procedure calls run at PASSIVE_LEVEL, not DISPATCH_LEVEL. ntoskrnl_run_dpc() has been fixed accordingly. (I read the documentation wrong.) - Similarly, the NDIS interrupt handler, which is essentially a DPC, also doesn't need to run at DISPATCH_LEVEL. ndis_intrtask() has been fixed accordingly. - MiniportQueryInformation() and MiniportSetInformation() run at DISPATCH_LEVEL, and each request must complete before another can be submitted. ndis_get_info() and ndis_set_info() have been fixed accordingly. - Turned the sleep lock that guards the NDIS thread job list into a spin lock. We never do anything with this lock held except manage the job list (no other locks are held), so it's safe to do this, and it's possible that ndis_sched() and ndis_unsched() can be called from DISPATCH_LEVEL, so using a sleep lock here is semantically incorrect. Also updated subr_witness.c to add the lock to the order list.
2005-03-07 03:05:31 +00:00
{ "NDIS thread lock", &lock_class_mtx_spin },
Register the machine check private state spinlock on ia64.
2002-10-12 00:33:36 +00:00
#endif
The latest release of the FreeBSD driver (twa) for 3ware's 9xxx series controllers. This corresponds to the 9.2 release (for FreeBSD 5.2.1) on the 3ware website. Highlights of this release are: 1. The driver has been re-architected to use a "Common Layer" (all tw_cl* files), which is a consolidation of all OS-independent parts of the driver. The FreeBSD OS specific portions of the driver go into an "OS Layer" (all tw_osl* files). This re-architecture is to achieve better maintainability, consistency of behavior across OS's, and better portability to new OS's (drivers for new OS's can be written by just adding an OS Layer that's specific to the OS, by complying to a "Common Layer Programming Interface" API. 2. The driver takes advantage of multiple processors. 3. The driver has a new firmware image bundled, the new features of which include Online Capacity Expansion and multi-lun support, among others. More details about 3ware's 9.2 release can be found here: http://www.3ware.com/download/Escalade9000Series/9.2/9.2_Release_Notes_Web.pdf Since the Common Layer is used across OS's, the FreeBSD specific include path for header files (/sys/dev/twa) is not part of the #include pre-processor directive in any of the source files. For being able to integrate twa into the kernel despite this, Makefile.<arch> has been changed to add the include path to CFLAGS. Reviewed by: scottl
2005-04-12 22:07:11 +00:00
{ "tw_osl_io_lock", &lock_class_mtx_spin },
{ "tw_osl_q_lock", &lock_class_mtx_spin },
{ "tw_cl_io_lock", &lock_class_mtx_spin },
{ "tw_cl_intr_lock", &lock_class_mtx_spin },
{ "tw_cl_gen_lock", &lock_class_mtx_spin },
Fix witness(4) warnings about mutex use. Group mutexes used in hwpmc(4) into 3 "types" in the sense of witness(4): - leaf spin mutexes---only one of these should be held at a time, so these mutexes are specified as belonging to a single witness type "pmc-leaf". - `struct pmc_owner' descriptors are protected by a spin mutex of witness type "pmc-owner-proc". Since we call wakeup_one() while holding these mutexes, the witness type of these mutexes needs to dominate that of "sleepq chain" mutexes. - logger threads use a sleep mutex, of type "pmc-sleep". Submitted by: wkoszek (earlier patch)
2007-04-19 08:02:51 +00:00
#ifdef HWPMC_HOOKS
{ "pmc-leaf", &lock_class_mtx_spin },
#endif
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
{ "blocked lock", &lock_class_mtx_spin },
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{ NULL, NULL },
{ NULL, NULL }
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
};
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#ifdef BLESSING
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
/*
* Pairs of locks which have been blessed
* Don't complain about order problems with blessed locks
*/
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
static struct witness_blessed blessed_list[] = {
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
};
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
static int blessed_count =
sizeof(blessed_list) / sizeof(struct witness_blessed);
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#endif
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* This global is set to 0 once it becomes safe to use the witness code.
*/
static int witness_cold = 1;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Don't panic if a spin lock is initialized that isn't in our static order list. Just warn about it instead. Requested by: scottl MFC after: 1 day
2005-10-24 20:14:24 +00:00
/*
* This global is set to 1 once the static lock orders have been enrolled
* so that a warning can be issued for any spin locks enrolled later.
*/
static int witness_spin_warn = 0;
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
/* Trim useless garbage from filenames. */
static const char *
fixup_filename(const char *file)
{
if (file == NULL)
return (NULL);
while (strncmp(file, "../", 3) == 0)
file += 3;
return (file);
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
* The WITNESS-enabled diagnostic code. Note that the witness code does
* assume that the early boot is single-threaded at least until after this
* routine is completed.
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
*/
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
static void
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
witness_initialize(void *dummy __unused)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct lock_object *lock;
struct witness_order_list_entry *order;
struct witness *w, *w1;
int i;
Fix a number of style issues in the MALLOC / FREE commit. I've tried to be careful not to fix anything that was already broken; the NFSv4 code is particularly bad in this respect.
2008-10-23 20:26:15 +00:00
w_data = malloc(sizeof (struct witness) * WITNESS_COUNT, M_WITNESS,
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
M_NOWAIT | M_ZERO);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* We have to release Giant before initializing its witness
* structure so that WITNESS doesn't get confused.
*/
mtx_unlock(&Giant);
mtx_assert(&Giant, MA_NOTOWNED);
Repeat after me -- "Use of ANSI string concatenation can be bad." In this case, C99's __func__ is properly defined as: static const char __func__[] = "function-name"; and GCC 3.1 will not allow it to be used in bogus string concatenation.
2001-12-10 05:40:12 +00:00
CTR1(KTR_WITNESS, "%s: initializing witness", __func__);
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
mtx_init(&w_mtx, "witness lock", NULL, MTX_SPIN | MTX_QUIET |
MUTEX_PROFILING has been generalized to LOCK_PROFILING. We now profile wait (time waited to acquire) and hold times for *all* kernel locks. If the architecture has a system synchronized TSC, the profiling code will use that - thereby minimizing profiling overhead. Large chunks of profiling code have been moved out of line, the overhead measured on the T1 for when it is compiled in but not enabled is < 1%. Approved by: scottl (standing in for mentor rwatson) Reviewed by: des and jhb
2006-11-11 03:18:07 +00:00
MTX_NOWITNESS | MTX_NOPROFILE);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
for (i = WITNESS_COUNT - 1; i >= 0; i--) {
w = &w_data[i];
memset(w, 0, sizeof(*w));
w_data[i].w_index = i; /* Witness index never changes. */
witness_free(w);
}
KASSERT(STAILQ_FIRST(&w_free)->w_index == 0,
("%s: Invalid list of free witness objects", __func__));
/* Witness with index 0 is not used to aid in debugging. */
STAILQ_REMOVE_HEAD(&w_free, w_list);
w_free_cnt--;
memset(w_rmatrix, 0,
(sizeof(**w_rmatrix) * (WITNESS_COUNT+1) * (WITNESS_COUNT+1)));
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
for (i = 0; i < LOCK_CHILDCOUNT; i++)
witness_lock_list_free(&w_locklistdata[i]);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_init_hash_tables();
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/* First add in all the specified order lists. */
for (order = order_lists; order->w_name != NULL; order++) {
w = enroll(order->w_name, order->w_class);
Check to see if enroll() returns NULL in the witness initialization. This can happen if witness runs out of resources during initialization or if witness_skipspin is enabled. Sleuthing by: Peter Jeremy <peter.jeremy@alcatel.com.au>
2001-04-17 03:35:38 +00:00
if (w == NULL)
continue;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w->w_file = "order list";
for (order++; order->w_name != NULL; order++) {
w1 = enroll(order->w_name, order->w_class);
Check to see if enroll() returns NULL in the witness initialization. This can happen if witness runs out of resources during initialization or if witness_skipspin is enabled. Sleuthing by: Peter Jeremy <peter.jeremy@alcatel.com.au>
2001-04-17 03:35:38 +00:00
if (w1 == NULL)
continue;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w1->w_file = "order list";
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
itismychild(w, w1);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w = w1;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
}
Don't panic if a spin lock is initialized that isn't in our static order list. Just warn about it instead. Requested by: scottl MFC after: 1 day
2005-10-24 20:14:24 +00:00
witness_spin_warn = 1;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/* Iterate through all locks and add them to witness. */
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
for (i = 0; pending_locks[i].wh_lock != NULL; i++) {
lock = pending_locks[i].wh_lock;
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
KASSERT(lock->lo_flags & LO_WITNESS,
("%s: lock %s is on pending list but not LO_WITNESS",
__func__, lock->lo_name));
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
lock->lo_witness = enroll(pending_locks[i].wh_type,
LOCK_CLASS(lock));
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
/* Mark the witness code as being ready for use. */
Don't panic if a spin lock is initialized that isn't in our static order list. Just warn about it instead. Requested by: scottl MFC after: 1 day
2005-10-24 20:14:24 +00:00
witness_cold = 0;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_lock(&Giant);
}
In keeping with style(9)'s recommendations on macros, use a ';' after each SYSINIT() macro invocation. This makes a number of lightweight C parsers much happier with the FreeBSD kernel source, including cflow's prcc and lxr. MFC after: 1 month Discussed with: imp, rink
2008-03-16 10:58:09 +00:00
SYSINIT(witness_init, SI_SUB_WITNESS, SI_ORDER_FIRST, witness_initialize,
NULL);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
void
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
witness_init(struct lock_object *lock, const char *type)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{
struct lock_class *class;
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
/* Various sanity checks. */
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if ((lock->lo_flags & LO_RECURSABLE) != 0 &&
(class->lc_flags & LC_RECURSABLE) == 0)
Style nits: - Don't use punctuation or newlines in panic messages. - Remove excess blank lines. Requested and partially submitted by: bde
2001-08-24 17:46:58 +00:00
panic("%s: lock (%s) %s can not be recursable", __func__,
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
class->lc_name, lock->lo_name);
if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
(class->lc_flags & LC_SLEEPABLE) == 0)
Style nits: - Don't use punctuation or newlines in panic messages. - Remove excess blank lines. Requested and partially submitted by: bde
2001-08-24 17:46:58 +00:00
panic("%s: lock (%s) %s can not be sleepable", __func__,
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
class->lc_name, lock->lo_name);
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
if ((lock->lo_flags & LO_UPGRADABLE) != 0 &&
(class->lc_flags & LC_UPGRADABLE) == 0)
Style nits: - Don't use punctuation or newlines in panic messages. - Remove excess blank lines. Requested and partially submitted by: bde
2001-08-24 17:46:58 +00:00
panic("%s: lock (%s) %s can not be upgradable", __func__,
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
class->lc_name, lock->lo_name);
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
/*
* If we shouldn't watch this lock, then just clear lo_witness.
* Otherwise, if witness_cold is set, then it is too early to
* enroll this lock, so defer it to witness_initialize() by adding
* it to the pending_locks list. If it is not too early, then enroll
* the lock now.
*/
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_watch < 1 || panicstr != NULL ||
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
(lock->lo_flags & LO_WITNESS) == 0)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
lock->lo_witness = NULL;
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
else if (witness_cold) {
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
pending_locks[pending_cnt].wh_lock = lock;
pending_locks[pending_cnt++].wh_type = type;
if (pending_cnt > WITNESS_PENDLIST)
panic("%s: pending locks list is too small, bump it\n",
__func__);
Tweak witness handling of lock object to shave 2 pointers off of each lock object (and thus off of each mutex and sx lock): - Rename the all_locks list to pending_locks and only put locks initialized before SI_SUB_WITNESS on the list so that the SI_SUB_WITNESS can add them to witness once it starts up. - Now that pending_locks is only used during early startup, change it from a TAILQ to an STAILQ. This removes a pointer from the STAILQ_ENTRY in struct lock_object. - Since the pending_locks list is only used during the single-threaded early boot it no longer needs to be protected by a mutex, so remove all_mtx. - Since the lo_list member of struct lock_object is now only used during early boot before witness is running, collapse lo_list and lo_witness into a union. This shaves the second pointer off of struct lock_object. - Axe lock_cur_cnt and lock_max_cnt. With these changes, struct mtx shrinks from 36 to 28 bytes on 32-bit platforms and from 72 to 56 bytes on 64-bit platforms. Note that this commit will completely and utterly destroy the kernel ABI, so no MFC. Tested on: alpha, amd64, i386, sparc64
2005-12-05 20:45:24 +00:00
} else
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
lock->lo_witness = enroll(type, class);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
void
witness_destroy(struct lock_object *lock)
{
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
struct lock_class *class;
Maintain a reference count on the witness struct. When the reference count drops to 0 in witness_destroy, set the w_name and w_file pointers to point to the string "(dead)" and the w_line field to 0. This way, if a mutex of a given name is used only in a module, then as long as all mutexes in the module are destroyed when the module is unloaded, witness will not maintain stale references to the mutex's name in the module's data section causing a panic later on when the w_name or w_file field's are examined.
2001-04-09 22:34:05 +00:00
struct witness *w;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (witness_cold)
panic("lock (%s) %s destroyed while witness_cold",
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class->lc_name, lock->lo_name);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
/* XXX: need to verify that no one holds the lock */
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if ((lock->lo_flags & LO_WITNESS) == 0 || lock->lo_witness == NULL)
return;
w = lock->lo_witness;
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
mtx_lock_spin(&w_mtx);
MPASS(w->w_refcount > 0);
w->w_refcount--;
if (w->w_refcount == 0)
depart(w);
mtx_unlock_spin(&w_mtx);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Adjust style of #ifdef's and #endif's to be more consistent and in line with recent additions to style(9).
2003-03-11 21:38:49 +00:00
#ifdef DDB
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
static void
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_compute_levels(void)
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness *w;
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
/*
* First clear all levels.
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
STAILQ_FOREACH(w, &w_all, w_list)
w->w_ddb_level = -1;
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* Look for locks with no parents and level all their descendants.
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
*/
STAILQ_FOREACH(w, &w_all, w_list) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* If the witness has ancestors (is not a root), skip it. */
if (w->w_num_ancestors > 0)
continue;
witness_ddb_level_descendants(w, 0);
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
}
}
static void
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_level_descendants(struct witness *w, int l)
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
{
int i;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (w->w_ddb_level >= l)
return;
w->w_ddb_level = l;
l++;
for (i = 1; i <= w_max_used_index; i++) {
if (w_rmatrix[w->w_index][i] & WITNESS_PARENT)
witness_ddb_level_descendants(&w_data[i], l);
}
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
}
static void
There is not a good reason to have a different prototype for db_printf() when compared to printf(). Unify it by returning the number of characters displayed for db_printf() as well. MFC after: 7 days
2010-05-11 17:01:14 +00:00
witness_ddb_display_descendants(int(*prnt)(const char *fmt, ...),
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness *w, int indent)
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
int i;
for (i = 0; i < indent; i++)
prnt(" ");
prnt("%s (type: %s, depth: %d, active refs: %d)",
w->w_name, w->w_class->lc_name,
w->w_ddb_level, w->w_refcount);
if (w->w_displayed) {
prnt(" -- (already displayed)\n");
return;
}
w->w_displayed = 1;
if (w->w_file != NULL && w->w_line != 0)
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
prnt(" -- last acquired @ %s:%d\n", fixup_filename(w->w_file),
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
w->w_line);
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
else
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
prnt(" -- never acquired\n");
indent++;
WITNESS_INDEX_ASSERT(w->w_index);
for (i = 1; i <= w_max_used_index; i++) {
if (w_rmatrix[w->w_index][i] & WITNESS_PARENT)
witness_ddb_display_descendants(prnt, &w_data[i],
indent);
Relocate witness_levelall(), witness_leveldescendents(), and witness_displaydescendants() so that they are protected by "#ifdef DDB/#endif" to unbreak kernels not using "option DDB". MFC after: 3 weeks
2005-09-11 07:57:06 +00:00
}
}
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
static void
There is not a good reason to have a different prototype for db_printf() when compared to printf(). Unify it by returning the number of characters displayed for db_printf() as well. MFC after: 7 days
2010-05-11 17:01:14 +00:00
witness_ddb_display_list(int(*prnt)(const char *fmt, ...),
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness_list *list)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
{
- Change witness_displaydescendants() to accept the indentation level as a parameter instead of using the level of a given witness. When recursing, pass an indent level of indent + 1. - Make use of the information witness_levelall() provides in witness_display_list() to use an O(n) algorithm instead of an O(n^2) algo to decide which witnesses to display hierarchies from. Basically, we only display a hierarchy for witnesses with a level of 0. - Add a new per-witness flag that is reset at the start of witness_display() for all witness's and is set the first time a witness is displayed in witness_displaydescendants(). If a witness is encountered more than once in the lock order tree (which happens often), witness_displaydescendants() marks the later occurrences with the string "(already displayed)" and doesn't display the subtree under that witness. This avoids duplicating large amounts of the lock order tree in the 'show witness' output in DDB. All these changes serve to make 'show witness' a lot more readable and useful than it was previously.
2003-03-11 22:14:21 +00:00
struct witness *w;
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
STAILQ_FOREACH(w, list, w_typelist) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (w->w_file == NULL || w->w_ddb_level > 0)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
continue;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* This lock has no anscestors - display its descendants. */
witness_ddb_display_descendants(prnt, w, 0);
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
Add a new witness sysctl which returns the relations between any lock and its children in the form: "parent","child" so that head and bottom of an oriented graph can be easilly detected and various form of diagrams can be build. The sysctl is called debug.witness.graphs and it is read-only; in order to get the list of relations, a simple: #sysctl debug.witness.graphs will do the trick. This approach has been choosen in order to support easilly things like the DOT format and such. Soon, an auto-explicative awk script, which filters simple informations returned by the sysctl and converts them into a real DOT script, will be committed to the repository between examples. Discussed with: rwatson
2008-05-07 21:41:36 +00:00
static void
There is not a good reason to have a different prototype for db_printf() when compared to printf(). Unify it by returning the number of characters displayed for db_printf() as well. MFC after: 7 days
2010-05-11 17:01:14 +00:00
witness_ddb_display(int(*prnt)(const char *fmt, ...))
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{
struct witness *w;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
witness_ddb_compute_levels();
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
- Change witness_displaydescendants() to accept the indentation level as a parameter instead of using the level of a given witness. When recursing, pass an indent level of indent + 1. - Make use of the information witness_levelall() provides in witness_display_list() to use an O(n) algorithm instead of an O(n^2) algo to decide which witnesses to display hierarchies from. Basically, we only display a hierarchy for witnesses with a level of 0. - Add a new per-witness flag that is reset at the start of witness_display() for all witness's and is set the first time a witness is displayed in witness_displaydescendants(). If a witness is encountered more than once in the lock order tree (which happens often), witness_displaydescendants() marks the later occurrences with the string "(already displayed)" and doesn't display the subtree under that witness. This avoids duplicating large amounts of the lock order tree in the 'show witness' output in DDB. All these changes serve to make 'show witness' a lot more readable and useful than it was previously.
2003-03-11 22:14:21 +00:00
/* Clear all the displayed flags. */
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
STAILQ_FOREACH(w, &w_all, w_list)
- Change witness_displaydescendants() to accept the indentation level as a parameter instead of using the level of a given witness. When recursing, pass an indent level of indent + 1. - Make use of the information witness_levelall() provides in witness_display_list() to use an O(n) algorithm instead of an O(n^2) algo to decide which witnesses to display hierarchies from. Basically, we only display a hierarchy for witnesses with a level of 0. - Add a new per-witness flag that is reset at the start of witness_display() for all witness's and is set the first time a witness is displayed in witness_displaydescendants(). If a witness is encountered more than once in the lock order tree (which happens often), witness_displaydescendants() marks the later occurrences with the string "(already displayed)" and doesn't display the subtree under that witness. This avoids duplicating large amounts of the lock order tree in the 'show witness' output in DDB. All these changes serve to make 'show witness' a lot more readable and useful than it was previously.
2003-03-11 22:14:21 +00:00
w->w_displayed = 0;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
- s/mutexes/locks/g in appropriate comments. - Rename the 'show mutexes' ddb command to 'show locks' since it shows a list of all the lock objects held by the current process.
2001-03-28 12:39:40 +00:00
* First, handle sleep locks which have been acquired at least
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
* once.
*/
prnt("Sleep locks:\n");
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_display_list(prnt, &w_sleep);
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
/*
- s/mutexes/locks/g in appropriate comments. - Rename the 'show mutexes' ddb command to 'show locks' since it shows a list of all the lock objects held by the current process.
2001-03-28 12:39:40 +00:00
* Now do spin locks which have been acquired at least once.
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
*/
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
prnt("\nSpin locks:\n");
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_display_list(prnt, &w_spin);
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
/*
- s/mutexes/locks/g in appropriate comments. - Rename the 'show mutexes' ddb command to 'show locks' since it shows a list of all the lock objects held by the current process.
2001-03-28 12:39:40 +00:00
* Finally, any locks which have not been acquired yet.
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
*/
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
prnt("\nLocks which were never acquired:\n");
STAILQ_FOREACH(w, &w_all, w_list) {
Handle "dead" witnesses better in the situation of several short term locks being created and destroyed without a single long-term one around to ensure the witness associated with that group of locks stays alive. The pipe mutexes are an example of this group. For a dead witness we no longer clear the witness name. Instead, when looking up the witness for a lock, if a dead witness' (a witness with a refcount of 0) w_name pointer is identical to the witness name of the lock then we revive that witness instead of using a new witness for the lock. This results in far fewer dead witness objects and also better preserves locking orders over the long term resulting in more correct lock order checking. Note that we can't ever derefence w_name of a dead witness since we don't know if the string it is pointing to has been free()'d or kldunload()'d out from under us.
2002-06-06 19:04:38 +00:00
if (w->w_file != NULL || w->w_refcount == 0)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
continue;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
prnt("%s (type: %s, depth: %d)\n", w->w_name,
w->w_class->lc_name, w->w_ddb_level);
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
}
}
Adjust style of #ifdef's and #endif's to be more consistent and in line with recent additions to style(9).
2003-03-11 21:38:49 +00:00
#endif /* DDB */
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
int
witness_defineorder(struct lock_object *lock1, struct lock_object *lock2)
{
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch == -1 || panicstr != NULL)
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
return (0);
/* Require locks that witness knows about. */
if (lock1 == NULL || lock1->lo_witness == NULL || lock2 == NULL ||
lock2->lo_witness == NULL)
return (EINVAL);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
mtx_assert(&w_mtx, MA_NOTOWNED);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
mtx_lock_spin(&w_mtx);
/*
* If we already have either an explicit or implied lock order that
* is the other way around, then return an error.
*/
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch &&
isitmydescendant(lock2->lo_witness, lock1->lo_witness)) {
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
mtx_unlock_spin(&w_mtx);
return (EDOOFUS);
}
/* Try to add the new order. */
CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__,
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
lock2->lo_witness->w_name, lock1->lo_witness->w_name);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
itismychild(lock1->lo_witness, lock2->lo_witness);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
mtx_unlock_spin(&w_mtx);
return (0);
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
void
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
witness_checkorder(struct lock_object *lock, int flags, const char *file,
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
int line, struct lock_object *interlock)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
struct lock_list_entry *lock_list, *lle;
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
struct lock_instance *lock1, *lock2, *plock;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct lock_class *class;
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
struct witness *w, *w1;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
int i, j;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || witness_watch < 1 || lock->lo_witness == NULL ||
- Fix panicstr checks to explicitly check against NULL. - Add a few more panicstr checks so that we don't panic recursively. Requested by: sheldonh (2)
2001-07-31 17:44:57 +00:00
panicstr != NULL)
- Add a new flag MTX_QUIET that can be passed to the various mtx_* functions. If this flag is set, then no KTR log messages are issued. This is useful for blocking excessive logging, such as with the internal mutex used by the witness code. - Use MTX_QUIET on all of the mtx_enter/exit operations on the internal mutex used by the witness code. - If we are in a panic, don't do witness checks in witness_enter(), witness_exit(), and witness_try_enter(), just return.
2000-12-13 21:53:42 +00:00
return;
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w = lock->lo_witness;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
td = curthread;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (class->lc_flags & LC_SLEEPLOCK) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Explicitly document how we implicitly enforce the lock order of sleep locks before spin locks.
2002-04-02 16:51:20 +00:00
/*
* Since spin locks include a critical section, this check
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
* implicitly enforces a lock order of all sleep locks before
Explicitly document how we implicitly enforce the lock order of sleep locks before spin locks.
2002-04-02 16:51:20 +00:00
* all spin locks.
*/
Don't "implicitly order all sleep locks before spin locks" in witness when the spin lock in question isn't -- it's the critical_enter() that KDB set. No more panic in DDB for console -> syscons -> tty -> knote operations.
2004-10-09 08:16:37 +00:00
if (td->td_critnest != 0 && !kdb_active)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
panic("blockable sleep lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
/*
* If this is the first lock acquired then just return as
* no order checking is needed.
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
lock_list = td->td_sleeplocks;
if (lock_list == NULL || lock_list->ll_count == 0)
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
return;
} else {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
/*
* If this is the first lock, just return as no order
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
* checking is needed. Avoid problems with thread
* migration pinning the thread while checking if
* spinlocks are held. If at least one spinlock is held
* the thread is in a safe path and it is allowed to
* unpin it.
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
sched_pin();
lock_list = PCPU_GET(spinlocks);
if (lock_list == NULL || lock_list->ll_count == 0) {
sched_unpin();
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
return;
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
}
sched_unpin();
Check the lock lists to see if they are empty directly rather than assigning a pointer to the list and then dereferencing the pointer as a second step. When the first spin lock is acquired, curthread is not in a critical section so it may be preempted and would end up using another CPUs lock list instead of its own. When this code was in witness_lock() this sequence was safe as curthread was in a critical section already since witness_lock() is called after the lock is acquired. Tested by: Daniel Lang dl at leo.org
2004-07-09 17:46:27 +00:00
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
/*
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
* Check to see if we are recursing on a lock we already own. If
* so, make sure that we don't mismatch exclusive and shared lock
* acquires.
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
lock1 = find_instance(lock_list, lock);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
if (lock1 != NULL) {
if ((lock1->li_flags & LI_EXCLUSIVE) != 0 &&
(flags & LOP_EXCLUSIVE) == 0) {
printf("shared lock of (%s) %s @ %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
printf("while exclusively locked from %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
fixup_filename(lock1->li_file), lock1->li_line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
panic("share->excl");
}
if ((lock1->li_flags & LI_EXCLUSIVE) == 0 &&
(flags & LOP_EXCLUSIVE) != 0) {
printf("exclusive lock of (%s) %s @ %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
printf("while share locked from %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
fixup_filename(lock1->li_file), lock1->li_line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
panic("excl->share");
}
return;
}
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
/*
* Find the previously acquired lock, but ignore interlocks.
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
plock = &lock_list->ll_children[lock_list->ll_count - 1];
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
if (interlock != NULL && plock->li_lock == interlock) {
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
if (lock_list->ll_count > 1)
plock =
&lock_list->ll_children[lock_list->ll_count - 2];
else {
lle = lock_list->ll_next;
- For any lock list we hold the head in order to reduce allocation from the free list and in this way avoid contention on the w_mtx. In order to make the code simple, we rely on the rule that when the head has not a child it also doesn't have other subsequent entries. Actually this assertion is broken because we can free all the head children and quit witness_unlock() with the head still allocated, with no children and subsequent entries present. Fix this by shifting the head if other entries are present and still freeing the object, but leaving always an head. - Fix witness_thread_has_locks() in order to report, correctly, if the lock list linked to a specific thread has children or not based on the above explained rule. - Fix a printout into DDB's "show alllocks" command in order to show, correctly, the process name that is really what we want. - Fix style(9) for a comment. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> Reported by: Marko Kiiskila <marko dot kiiskila at nokia dot com> Sponsored by: Nokia
2008-09-12 21:44:01 +00:00
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
/*
* The interlock is the only lock we hold, so
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
* simply return.
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
if (lle == NULL)
return;
plock = &lle->ll_children[lle->ll_count - 1];
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
}
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Try to perform most checks without a lock. If this succeeds we
* can skip acquiring the lock and return success.
*/
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
w1 = plock->li_lock->lo_witness;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (witness_lock_order_check(w1, w))
return;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* Check for duplicate locks of the same type. Note that we only
* have to check for this on the last lock we just acquired. Any
* other cases will be caught as lock order violations.
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
mtx_lock_spin(&w_mtx);
witness_lock_order_add(w1, w);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (w1 == w) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
i = w->w_index;
if (!(lock->lo_flags & LO_DUPOK) && !(flags & LOP_DUPOK) &&
!(w_rmatrix[i][i] & WITNESS_REVERSAL)) {
w_rmatrix[i][i] |= WITNESS_REVERSAL;
w->w_reversed = 1;
mtx_unlock_spin(&w_mtx);
Oops, missed updating a place with with 's/lock1/plock/' when adding interlock support to WITNESS. Specifically, the printf listing the first location when duplicate locks of the same type are acquired. Reported by: pho
2008-10-03 18:13:05 +00:00
printf(
"acquiring duplicate lock of same type: \"%s\"\n",
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
w->w_name);
Oops, missed updating a place with with 's/lock1/plock/' when adding interlock support to WITNESS. Specifically, the printf listing the first location when duplicate locks of the same type are acquired. Reported by: pho
2008-10-03 18:13:05 +00:00
printf(" 1st %s @ %s:%d\n", plock->li_lock->lo_name,
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
fixup_filename(plock->li_file), plock->li_line);
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
printf(" 2nd %s @ %s:%d\n", lock->lo_name,
fixup_filename(file), line);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_debugger(1);
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
} else
mtx_unlock_spin(&w_mtx);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
return;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
mtx_assert(&w_mtx, MA_OWNED);
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
/*
Fix typos - remove duplicate "the". PR: bin/154928 Submitted by: Eitan Adler <lists at eitanadler.com> MFC after: 3 days
2011-02-21 09:01:34 +00:00
* If we know that the lock we are acquiring comes after
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
* the lock we most recently acquired in the lock order tree,
* then there is no need for any further checks.
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (isitmychild(w1, w))
goto out;
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
for (j = 0, lle = lock_list; lle != NULL; lle = lle->ll_next) {
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
for (i = lle->ll_count - 1; i >= 0; i--, j++) {
MPASS(j < WITNESS_COUNT);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
lock1 = &lle->ll_children[i];
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
/*
* Ignore the interlock the first time we see it.
*/
if (interlock != NULL && interlock == lock1->li_lock) {
interlock = NULL;
continue;
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* If this lock doesn't undergo witness checking,
* then skip it.
*/
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
w1 = lock1->li_lock->lo_witness;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (w1 == NULL) {
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
KASSERT((lock1->li_lock->lo_flags & LO_WITNESS) == 0,
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
("lock missing witness structure"));
continue;
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
/*
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
* If we are locking Giant and this is a sleepable
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
* lock, then skip it.
*/
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0 &&
Rename the 'mtx_object', 'rw_object', and 'sx_object' members of mutexes, rwlocks, and sx locks to 'lock_object'.
2007-03-21 21:20:51 +00:00
lock == &Giant.lock_object)
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
continue;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Enforce an implicit lock order of sleepable locks before non-sleepable locks.
2002-04-02 19:27:21 +00:00
/*
* If we are locking a sleepable lock and this lock
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
* is Giant, then skip it.
*/
if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
Rename the 'mtx_object', 'rw_object', and 'sx_object' members of mutexes, rwlocks, and sx locks to 'lock_object'.
2007-03-21 21:20:51 +00:00
lock1->li_lock == &Giant.lock_object)
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
continue;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
/*
* If we are locking a sleepable lock and this lock
* isn't sleepable, we want to treat it as a lock
* order violation to enfore a general lock order of
* sleepable locks before non-sleepable locks.
Enforce an implicit lock order of sleepable locks before non-sleepable locks.
2002-04-02 19:27:21 +00:00
*/
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
if (((lock->lo_flags & LO_SLEEPABLE) != 0 &&
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
(lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0))
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
goto reversal;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
- Enforce an implicit lock order that Giant cannot be locked while holding any other non-sleepable lock. In plain English: Giant comes before all other mutexes. - Add some extra description to the lock order reversal printf's to indicate when a reversal is triggered by a hard-coded implicit rule. Requested by: truckman (2) MFC after: 1 week
2005-09-15 19:07:14 +00:00
/*
* If we are locking Giant and this is a non-sleepable
* lock, then treat it as a reversal.
*/
if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0 &&
Rename the 'mtx_object', 'rw_object', and 'sx_object' members of mutexes, rwlocks, and sx locks to 'lock_object'.
2007-03-21 21:20:51 +00:00
lock == &Giant.lock_object)
- Enforce an implicit lock order that Giant cannot be locked while holding any other non-sleepable lock. In plain English: Giant comes before all other mutexes. - Add some extra description to the lock order reversal printf's to indicate when a reversal is triggered by a hard-coded implicit rule. Requested by: truckman (2) MFC after: 1 week
2005-09-15 19:07:14 +00:00
goto reversal;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
/*
* Check the lock order hierarchy for a reveresal.
*/
if (!isitmydescendant(w, w1))
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
continue;
- Add some comments to some of the static lock orders. Don't explicitly link proctree and allproc to Giant since that order is already implicitly enforced. - Use a goto to handle the case where we want to enforce a reversal before calling isitmydescendant() in witness_checkorder() so that the logic is easier to follow and so that it is easier to add more forced-reversal cases in the future. MFC after: 3 days
2005-09-02 20:23:49 +00:00
reversal:
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* We have a lock order violation, check to see if it
* is allowed or has already been yelled about.
*/
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#ifdef BLESSING
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
/*
* If the lock order is blessed, just bail. We don't
* look for other lock order violations though, which
* may be a bug.
*/
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
if (blessed(w, w1))
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
goto out;
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#endif
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* Bail if this violation is known */
if (w_rmatrix[w1->w_index][w->w_index] & WITNESS_REVERSAL)
goto out;
/* Record this as a violation */
w_rmatrix[w1->w_index][w->w_index] |= WITNESS_REVERSAL;
w_rmatrix[w->w_index][w1->w_index] |= WITNESS_REVERSAL;
w->w_reversed = w1->w_reversed = 1;
witness_increment_graph_generation();
mtx_unlock_spin(&w_mtx);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* Ok, yell about it.
*/
- Enforce an implicit lock order that Giant cannot be locked while holding any other non-sleepable lock. In plain English: Giant comes before all other mutexes. - Add some extra description to the lock order reversal printf's to indicate when a reversal is triggered by a hard-coded implicit rule. Requested by: truckman (2) MFC after: 1 week
2005-09-15 19:07:14 +00:00
if (((lock->lo_flags & LO_SLEEPABLE) != 0 &&
(lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0))
printf(
"lock order reversal: (sleepable after non-sleepable)\n");
else if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0
Rename the 'mtx_object', 'rw_object', and 'sx_object' members of mutexes, rwlocks, and sx locks to 'lock_object'.
2007-03-21 21:20:51 +00:00
&& lock == &Giant.lock_object)
- Enforce an implicit lock order that Giant cannot be locked while holding any other non-sleepable lock. In plain English: Giant comes before all other mutexes. - Add some extra description to the lock order reversal printf's to indicate when a reversal is triggered by a hard-coded implicit rule. Requested by: truckman (2) MFC after: 1 week
2005-09-15 19:07:14 +00:00
printf(
"lock order reversal: (Giant after non-sleepable)\n");
else
printf("lock order reversal:\n");
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
* Try to locate an earlier lock with
* witness w in our list.
*/
do {
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
lock2 = &lle->ll_children[i];
MPASS(lock2->li_lock != NULL);
if (lock2->li_lock->lo_witness == w)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
break;
if (i == 0 && lle->ll_next != NULL) {
lle = lle->ll_next;
i = lle->ll_count - 1;
Correct an assertion in the code to traverse the list of locks to find an earlier acquired lock with the same witness as the lock currently being acquired. If we had released several earlier acquired locks after acquiring enough locks to require another lock_list_entry bucket in the lock list, then subsequent lock_list_entry buckets could contain only one lock instance in which case i would be zero. Reported by: Joel M. Baldwin <qumqats@outel.org>
2002-11-11 16:36:20 +00:00
MPASS(i >= 0 && i < LOCK_NCHILDREN);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
} else
i--;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
} while (i >= 0);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
if (i < 0) {
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
printf(" 1st %p %s (%s) @ %s:%d\n",
lock1->li_lock, lock1->li_lock->lo_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
w1->w_name, fixup_filename(lock1->li_file),
lock1->li_line);
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
printf(" 2nd %p %s (%s) @ %s:%d\n", lock,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
lock->lo_name, w->w_name,
fixup_filename(file), line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
} else {
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
printf(" 1st %p %s (%s) @ %s:%d\n",
lock2->li_lock, lock2->li_lock->lo_name,
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
lock2->li_lock->lo_witness->w_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
fixup_filename(lock2->li_file),
lock2->li_line);
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
printf(" 2nd %p %s (%s) @ %s:%d\n",
lock1->li_lock, lock1->li_lock->lo_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
w1->w_name, fixup_filename(lock1->li_file),
lock1->li_line);
Add a new char * pointer lo_type to struct lock_object that is used to point to a more generic name for a lock that is more suitable for use by witness when grouping locks. For example, although network driver locks use the interface name for the name of each lock, they should all use the same witness and be treated the same as witness. Another example is that all UMA zone locks should be treated the same. The witness code has also been updated to print out the lock type in addition to the lock name in a few places where it is relevant.
2002-04-04 20:45:21 +00:00
printf(" 3rd %p %s (%s) @ %s:%d\n", lock,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
lock->lo_name, w->w_name,
fixup_filename(file), line);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_debugger(1);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
return;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
/*
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
* If requested, build a new lock order. However, don't build a new
* relationship between a sleepable lock and Giant if it is in the
* wrong direction. The correct lock order is that sleepable locks
* always come before Giant.
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
*/
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
if (flags & LOP_NEWORDER &&
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
!(plock->li_lock == &Giant.lock_object &&
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
(lock->lo_flags & LO_SLEEPABLE) != 0)) {
Repeat after me -- "Use of ANSI string concatenation can be bad." In this case, C99's __func__ is properly defined as: static const char __func__[] = "function-name"; and GCC 3.1 will not allow it to be used in bogus string concatenation.
2001-12-10 05:40:12 +00:00
CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__,
Teach WITNESS about the interlocks used with lockmgr. This removes a bunch of spurious witness warnings since lockmgr grew witness support. Before this, every time you passed an interlock to a lockmgr lock WITNESS treated it as a LOR. Reviewed by: attilio
2008-09-10 19:13:30 +00:00
w->w_name, plock->li_lock->lo_witness->w_name);
itismychild(plock->li_lock->lo_witness, w);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
}
out:
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
mtx_unlock_spin(&w_mtx);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
}
void
witness_lock(struct lock_object *lock, int flags, const char *file, int line)
{
struct lock_list_entry **lock_list, *lle;
struct lock_instance *instance;
struct witness *w;
struct thread *td;
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || witness_watch == -1 || lock->lo_witness == NULL ||
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
panicstr != NULL)
return;
w = lock->lo_witness;
td = curthread;
/* Determine lock list for this lock. */
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
if (LOCK_CLASS(lock)->lc_flags & LC_SLEEPLOCK)
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
lock_list = &td->td_sleeplocks;
else
lock_list = PCPU_PTR(spinlocks);
/* Check to see if we are recursing on a lock we already own. */
instance = find_instance(*lock_list, lock);
if (instance != NULL) {
instance->li_flags++;
CTR4(KTR_WITNESS, "%s: pid %d recursed on %s r=%d", __func__,
td->td_proc->p_pid, lock->lo_name,
instance->li_flags & LI_RECURSEMASK);
instance->li_file = file;
instance->li_line = line;
return;
}
/* Update per-witness last file and line acquire. */
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
w->w_file = file;
w->w_line = line;
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
/* Find the next open lock instance in the list and fill it. */
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
lle = *lock_list;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
if (lle == NULL || lle->ll_count == LOCK_NCHILDREN) {
- Move the 'clk' spinlock below other spin locks since KTR trace events may need the clock lock for nanotime(). - Add KTR trace events for lock list manipulations and other witness operations. - Use a temporary variable instead of setting the lock list head directly and then setting up the links to add a new lock list entry to the lock list. This small race could result in witness "forgetting" about all the locks held by this process temporarily during an interrupt. - Close a more fatal race condition when removing a lock from a list. Removing a lock from the list entails both decrementing the count of items in this bucket as well as shuffling items in the current bucket up a notch to replace the gap left by the removed item. Wrap these operations in a critical section.
2001-06-25 23:17:52 +00:00
lle = witness_lock_list_get();
if (lle == NULL)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return;
- Move the 'clk' spinlock below other spin locks since KTR trace events may need the clock lock for nanotime(). - Add KTR trace events for lock list manipulations and other witness operations. - Use a temporary variable instead of setting the lock list head directly and then setting up the links to add a new lock list entry to the lock list. This small race could result in witness "forgetting" about all the locks held by this process temporarily during an interrupt. - Close a more fatal race condition when removing a lock from a list. Removing a lock from the list entails both decrementing the count of items in this bucket as well as shuffling items in the current bucket up a notch to replace the gap left by the removed item. Wrap these operations in a critical section.
2001-06-25 23:17:52 +00:00
lle->ll_next = *lock_list;
Repeat after me -- "Use of ANSI string concatenation can be bad." In this case, C99's __func__ is properly defined as: static const char __func__[] = "function-name"; and GCC 3.1 will not allow it to be used in bogus string concatenation.
2001-12-10 05:40:12 +00:00
CTR3(KTR_WITNESS, "%s: pid %d added lle %p", __func__,
Replace 'curproc' with 'td->td_proc'.
2001-10-08 21:05:46 +00:00
td->td_proc->p_pid, lle);
- Move the 'clk' spinlock below other spin locks since KTR trace events may need the clock lock for nanotime(). - Add KTR trace events for lock list manipulations and other witness operations. - Use a temporary variable instead of setting the lock list head directly and then setting up the links to add a new lock list entry to the lock list. This small race could result in witness "forgetting" about all the locks held by this process temporarily during an interrupt. - Close a more fatal race condition when removing a lock from a list. Removing a lock from the list entails both decrementing the count of items in this bucket as well as shuffling items in the current bucket up a notch to replace the gap left by the removed item. Wrap these operations in a critical section.
2001-06-25 23:17:52 +00:00
*lock_list = lle;
Implement MTX_RECURSE flag for mtx_init(). All calls to mtx_init() for mutexes that recurse must now include the MTX_RECURSE bit in the flag argument variable. This change is in preparation for an upcoming (further) mutex API cleanup. The witness code will call panic() if a lock is found to recurse but the MTX_RECURSE bit was not set during the lock's initialization. The old MTX_RECURSE "state" bit (in mtx_lock) has been renamed to MTX_RECURSED, which is more appropriate given its meaning. The following locks have been made "recursive," thus far: eventhandler, Giant, callout, sched_lock, possibly some others declared in the architecture-specific code, all of the network card driver locks in pci/, as well as some other locks in dev/ stuff that I've found to be recursive. Reviewed by: jhb
2001-01-19 01:59:14 +00:00
}
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
instance = &lle->ll_children[lle->ll_count++];
instance->li_lock = lock;
instance->li_line = line;
instance->li_file = file;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
if ((flags & LOP_EXCLUSIVE) != 0)
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
instance->li_flags = LI_EXCLUSIVE;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
else
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
instance->li_flags = 0;
Repeat after me -- "Use of ANSI string concatenation can be bad." In this case, C99's __func__ is properly defined as: static const char __func__[] = "function-name"; and GCC 3.1 will not allow it to be used in bogus string concatenation.
2001-12-10 05:40:12 +00:00
CTR4(KTR_WITNESS, "%s: pid %d added %s as lle[%d]", __func__,
Replace 'curproc' with 'td->td_proc'.
2001-10-08 21:05:46 +00:00
td->td_proc->p_pid, lock->lo_name, lle->ll_count - 1);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
void
witness_upgrade(struct lock_object *lock, int flags, const char *file, int line)
{
struct lock_instance *instance;
struct lock_class *class;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (lock->lo_witness == NULL || witness_watch == -1 || panicstr != NULL)
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
return;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch) {
if ((lock->lo_flags & LO_UPGRADABLE) == 0)
panic("upgrade of non-upgradable lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if ((class->lc_flags & LC_SLEEPLOCK) == 0)
panic("upgrade of non-sleep lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
}
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
instance = find_instance(curthread->td_sleeplocks, lock);
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
if (instance == NULL)
panic("upgrade of unlocked lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch) {
if ((instance->li_flags & LI_EXCLUSIVE) != 0)
panic("upgrade of exclusive lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if ((instance->li_flags & LI_RECURSEMASK) != 0)
panic("upgrade of recursed lock (%s) %s r=%d @ %s:%d",
class->lc_name, lock->lo_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
instance->li_flags & LI_RECURSEMASK,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
}
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
instance->li_flags |= LI_EXCLUSIVE;
}
void
witness_downgrade(struct lock_object *lock, int flags, const char *file,
int line)
{
struct lock_instance *instance;
struct lock_class *class;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (lock->lo_witness == NULL || witness_watch == -1 || panicstr != NULL)
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
return;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch) {
if ((lock->lo_flags & LO_UPGRADABLE) == 0)
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
panic("downgrade of non-upgradable lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if ((class->lc_flags & LC_SLEEPLOCK) == 0)
panic("downgrade of non-sleep lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
}
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
instance = find_instance(curthread->td_sleeplocks, lock);
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
if (instance == NULL)
panic("downgrade of unlocked lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch) {
if ((instance->li_flags & LI_EXCLUSIVE) == 0)
panic("downgrade of shared lock (%s) %s @ %s:%d",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if ((instance->li_flags & LI_RECURSEMASK) != 0)
panic("downgrade of recursed lock (%s) %s r=%d @ %s:%d",
class->lc_name, lock->lo_name,
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
instance->li_flags & LI_RECURSEMASK,
fixup_filename(file), line);
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
}
Add witness_upgrade() and witness_downgrade() for handling upgrades and downgrades of shared/exclusive locks.
2001-08-23 22:47:05 +00:00
instance->li_flags &= ~LI_EXCLUSIVE;
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
void
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
witness_unlock(struct lock_object *lock, int flags, const char *file, int line)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct lock_list_entry **lock_list, *lle;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
struct lock_instance *instance;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct lock_class *class;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td;
Remove last two abuses of cpu_critical_{enter,exit} in the MI code. Reviewed by: jake, jhb, rwatson
2002-03-21 06:11:09 +00:00
register_t s;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
int i, j;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || lock->lo_witness == NULL || panicstr != NULL)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
return;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
td = curthread;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
/* Find lock instance associated with this lock. */
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
if (class->lc_flags & LC_SLEEPLOCK)
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
lock_list = &td->td_sleeplocks;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
else
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
lock_list = PCPU_PTR(spinlocks);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
lle = *lock_list;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
for (; *lock_list != NULL; lock_list = &(*lock_list)->ll_next)
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
for (i = 0; i < (*lock_list)->ll_count; i++) {
instance = &(*lock_list)->ll_children[i];
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
if (instance->li_lock == lock)
goto found;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
}
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
/*
* When disabling WITNESS through witness_watch we could end up in
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
* having registered locks in the td_sleeplocks queue.
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
* We have to make sure we flush these queues, so just search for
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
* eventual register locks and remove them.
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
*/
if (witness_watch > 0)
panic("lock (%s) %s not locked @ %s:%d", class->lc_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
lock->lo_name, fixup_filename(file), line);
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
else
return;
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
found:
/* First, check for shared/exclusive mismatches. */
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if ((instance->li_flags & LI_EXCLUSIVE) != 0 && witness_watch > 0 &&
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
(flags & LOP_EXCLUSIVE) == 0) {
printf("shared unlock of (%s) %s @ %s:%d\n", class->lc_name,
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
lock->lo_name, fixup_filename(file), line);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
printf("while exclusively locked from %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
fixup_filename(instance->li_file), instance->li_line);
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
panic("excl->ushare");
}
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if ((instance->li_flags & LI_EXCLUSIVE) == 0 && witness_watch > 0 &&
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
(flags & LOP_EXCLUSIVE) != 0) {
printf("exclusive unlock of (%s) %s @ %s:%d\n", class->lc_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
lock->lo_name, fixup_filename(file), line);
printf("while share locked from %s:%d\n",
fixup_filename(instance->li_file),
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
instance->li_line);
panic("share->uexcl");
}
/* If we are recursed, unrecurse. */
if ((instance->li_flags & LI_RECURSEMASK) > 0) {
CTR4(KTR_WITNESS, "%s: pid %d unrecursed on %s r=%d", __func__,
td->td_proc->p_pid, instance->li_lock->lo_name,
instance->li_flags);
instance->li_flags--;
return;
}
Move the NORELEASE check to after the recurse count decrement and bailout, this is not counted as actually releasing the lock.
2009-02-28 19:10:43 +00:00
/* The lock is now being dropped, check for NORELEASE flag */
if ((instance->li_flags & LI_NORELEASE) != 0 && witness_watch > 0) {
printf("forbidden unlock of (%s) %s @ %s:%d\n", class->lc_name,
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
lock->lo_name, fixup_filename(file), line);
Move the NORELEASE check to after the recurse count decrement and bailout, this is not counted as actually releasing the lock.
2009-02-28 19:10:43 +00:00
panic("lock marked norelease");
}
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
/* Otherwise, remove this item from the list. */
s = intr_disable();
CTR4(KTR_WITNESS, "%s: pid %d removed %s from lle[%d]", __func__,
td->td_proc->p_pid, instance->li_lock->lo_name,
(*lock_list)->ll_count - 1);
for (j = i; j < (*lock_list)->ll_count - 1; j++)
(*lock_list)->ll_children[j] =
(*lock_list)->ll_children[j + 1];
(*lock_list)->ll_count--;
intr_restore(s);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
- For any lock list we hold the head in order to reduce allocation from the free list and in this way avoid contention on the w_mtx. In order to make the code simple, we rely on the rule that when the head has not a child it also doesn't have other subsequent entries. Actually this assertion is broken because we can free all the head children and quit witness_unlock() with the head still allocated, with no children and subsequent entries present. Fix this by shifting the head if other entries are present and still freeing the object, but leaving always an head. - Fix witness_thread_has_locks() in order to report, correctly, if the lock list linked to a specific thread has children or not based on the above explained rule. - Fix a printout into DDB's "show alllocks" command in order to show, correctly, the process name that is really what we want. - Fix style(9) for a comment. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> Reported by: Marko Kiiskila <marko dot kiiskila at nokia dot com> Sponsored by: Nokia
2008-09-12 21:44:01 +00:00
* In order to reduce contention on w_mtx, we want to keep always an
* head object into lists so that frequent allocation from the
* free witness pool (and subsequent locking) is avoided.
* In order to maintain the current code simple, when the head
* object is totally unloaded it means also that we do not have
* further objects in the list, so the list ownership needs to be
* hand over to another object if the current head needs to be freed.
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
*/
- For any lock list we hold the head in order to reduce allocation from the free list and in this way avoid contention on the w_mtx. In order to make the code simple, we rely on the rule that when the head has not a child it also doesn't have other subsequent entries. Actually this assertion is broken because we can free all the head children and quit witness_unlock() with the head still allocated, with no children and subsequent entries present. Fix this by shifting the head if other entries are present and still freeing the object, but leaving always an head. - Fix witness_thread_has_locks() in order to report, correctly, if the lock list linked to a specific thread has children or not based on the above explained rule. - Fix a printout into DDB's "show alllocks" command in order to show, correctly, the process name that is really what we want. - Fix style(9) for a comment. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> Reported by: Marko Kiiskila <marko dot kiiskila at nokia dot com> Sponsored by: Nokia
2008-09-12 21:44:01 +00:00
if ((*lock_list)->ll_count == 0) {
if (*lock_list == lle) {
if (lle->ll_next == NULL)
return;
} else
lle = *lock_list;
Rework witness_lock() to make it slightly more useful and flexible. - witness_lock() is split into two pieces: witness_checkorder() and witness_lock(). Witness_checkorder() determines if acquiring a specified lock at the time it is called would result in a lock order. It optionally adds a new lock order relationship as well. witness_lock() updates witness's data structures to assume that a lock has been acquired by stick a new lock instance in the appropriate lock instance list. - The mutex and sx lock functions now call checkorder() prior to trying to acquire a lock and continue to call witness_lock() after the acquire is completed. This will let witness catch a deadlock before it happens rather than trying to do so after the threads have deadlocked (i.e. never actually report it). - A new function witness_defineorder() has been added that adds a lock order between two locks at runtime without having to acquire the locks. If the lock order cannot be added it will return an error. This function is available to programmers via the WITNESS_DEFINEORDER() macro which accepts either two mutexes or two sx locks as its arguments. - A few simple wrapper macros were added to allow developers to call witness_checkorder() anywhere as a way of enforcing locking assertions in code that might acquire a certain lock in some situations. The macros are: witness_check_{mutex,shared_sx,exclusive_sx} and take an appropriate lock as the sole argument. - The code to remove a lock instance from a lock list in witness_unlock() was unnested by using a goto to vastly improve the readability of this function.
2004-01-28 20:39:57 +00:00
*lock_list = lle->ll_next;
CTR3(KTR_WITNESS, "%s: pid %d removed lle %p", __func__,
td->td_proc->p_pid, lle);
witness_lock_list_free(lle);
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
void
witness_thread_exit(struct thread *td)
{
struct lock_list_entry *lle;
int i, n;
lle = td->td_sleeplocks;
if (lle == NULL || panicstr != NULL)
return;
if (lle->ll_count != 0) {
for (n = 0; lle != NULL; lle = lle->ll_next)
for (i = lle->ll_count - 1; i >= 0; i--) {
if (n == 0)
printf("Thread %p exiting with the following locks held:\n",
td);
n++;
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_lock(&lle->ll_children[i], printf);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
}
panic("Thread %p cannot exit while holding sleeplocks\n", td);
}
witness_lock_list_free(lle);
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
/*
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
* Warn if any locks other than 'lock' are held. Flags can be passed in to
* exempt Giant and sleepable locks from the checks as well. If any
* non-exempt locks are held, then a supplied message is printed to the
* console along with a list of the offending locks. If indicated in the
* flags then a failure results in a panic as well.
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
*/
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
int
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
witness_warn(int flags, struct lock_object *lock, const char *fmt, ...)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
struct lock_list_entry *lock_list, *lle;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
struct lock_instance *lock1;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td;
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
va_list ap;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
int i, n;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || witness_watch < 1 || panicstr != NULL)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
return (0);
n = 0;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
td = curthread;
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
for (lle = td->td_sleeplocks; lle != NULL; lle = lle->ll_next)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
for (i = lle->ll_count - 1; i >= 0; i--) {
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
lock1 = &lle->ll_children[i];
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
if (lock1->li_lock == lock)
continue;
if (flags & WARN_GIANTOK &&
Rename the 'mtx_object', 'rw_object', and 'sx_object' members of mutexes, rwlocks, and sx locks to 'lock_object'.
2007-03-21 21:20:51 +00:00
lock1->li_lock == &Giant.lock_object)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
continue;
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
if (flags & WARN_SLEEPOK &&
(lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0)
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
continue;
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
if (n == 0) {
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
printf(" with the following");
if (flags & WARN_SLEEPOK)
printf(" non-sleepable");
Insert cosmetic spaces. Reported by: kris
2003-08-04 19:24:25 +00:00
printf(" locks held:\n");
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
n++;
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_lock(lock1, printf);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
/*
* Pin the thread in order to avoid problems with thread migration.
* Once that all verifies are passed about spinlocks ownership,
* the thread is in a safe path and it can be unpinned.
*/
sched_pin();
lock_list = PCPU_GET(spinlocks);
In the actual code for witness_warn: - If there aren't spinlocks held, but there are problems with old sleeplocks, they are not reported. - If the spinlock found is not the only one, problems are not reported. Fix these 2 problems. Reported by: tegge
2008-10-20 19:22:16 +00:00
if (lock_list != NULL && lock_list->ll_count != 0) {
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
sched_unpin();
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
- Allow witness_sleep() to be called when witness hasn't been initialized yet. We just return without performing any checks. - Don't explicitly enter and exit critical sections when walking lock lists. We don't need a critical section to walk the list of sleep locks for a thread. We check to see if a spin lock list is empty before we walk it. If the list is empty we don't need to walk it. If it isn't then we already hold at least one spin lock and are already in a critical section and thus don't need our own explicit critical section.
2002-05-20 17:49:46 +00:00
/*
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
* We should only have one spinlock and as long as
* the flags cannot match for this locks class,
* check if the first spinlock is the one curthread
* should hold.
- Allow witness_sleep() to be called when witness hasn't been initialized yet. We just return without performing any checks. - Don't explicitly enter and exit critical sections when walking lock lists. We don't need a critical section to walk the list of sleep locks for a thread. We check to see if a spin lock list is empty before we walk it. If the list is empty we don't need to walk it. If it isn't then we already hold at least one spin lock and are already in a critical section and thus don't need our own explicit critical section.
2002-05-20 17:49:46 +00:00
*/
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
lock1 = &lock_list->ll_children[lock_list->ll_count - 1];
In the actual code for witness_warn: - If there aren't spinlocks held, but there are problems with old sleeplocks, they are not reported. - If the spinlock found is not the only one, problems are not reported. Fix these 2 problems. Reported by: tegge
2008-10-20 19:22:16 +00:00
if (lock_list->ll_count == 1 && lock_list->ll_next == NULL &&
lock1->li_lock == lock && n == 0)
return (0);
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
printf(" with the following");
if (flags & WARN_SLEEPOK)
printf(" non-sleepable");
printf(" locks held:\n");
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
n += witness_list_locks(&lock_list, printf);
- Fix a race in witness_checkorder() where, between the PCPU_GET() and PCPU_PTR() curthread can migrate on another CPU and get incorrect results. - Fix a similar race into witness_warn(). - Fix the interlock's checks bypassing by correctly using the appropriate children even when the lock_list chunk to be explored is not the first one. - Allow witness_warn() to work with spinlocks too. Bugs found by: tegge Submitted by: jhb, tegge Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-16 12:42:56 +00:00
} else
sched_unpin();
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
if (flags & WARN_PANIC && n)
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
panic("%s", __func__);
else
witness_debugger(n);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (n);
}
Add WITNESS_FILE() and WITNESS_LINE(), which allow users of witness to print out the file and line from the lock object. These will be used shortly by CTR() calls in the mutex code. Reviewed by: jhb, jake
2002-08-26 18:31:26 +00:00
const char *
witness_file(struct lock_object *lock)
{
struct witness *w;
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || witness_watch < 1 || lock->lo_witness == NULL)
Add WITNESS_FILE() and WITNESS_LINE(), which allow users of witness to print out the file and line from the lock object. These will be used shortly by CTR() calls in the mutex code. Reviewed by: jhb, jake
2002-08-26 18:31:26 +00:00
return ("?");
w = lock->lo_witness;
return (w->w_file);
}
int
witness_line(struct lock_object *lock)
{
struct witness *w;
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_cold || witness_watch < 1 || lock->lo_witness == NULL)
Add WITNESS_FILE() and WITNESS_LINE(), which allow users of witness to print out the file and line from the lock object. These will be used shortly by CTR() calls in the mutex code. Reviewed by: jhb, jake
2002-08-26 18:31:26 +00:00
return (0);
w = lock->lo_witness;
return (w->w_line);
}
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
static struct witness *
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
enroll(const char *description, struct lock_class *lock_class)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct witness *w;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
struct witness_list *typelist;
MPASS(description != NULL);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch == -1 || panicstr != NULL)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (NULL);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if ((lock_class->lc_flags & LC_SPINLOCK)) {
if (witness_skipspin)
return (NULL);
else
typelist = &w_spin;
} else if ((lock_class->lc_flags & LC_SLEEPLOCK))
typelist = &w_sleep;
else
panic("lock class %s is not sleep or spin",
lock_class->lc_name);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_lock_spin(&w_mtx);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
w = witness_hash_get(description);
if (w)
goto found;
if ((w = witness_get()) == NULL)
return (NULL);
MPASS(strlen(description) < MAX_W_NAME);
strcpy(w->w_name, description);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w->w_class = lock_class;
Maintain a reference count on the witness struct. When the reference count drops to 0 in witness_destroy, set the w_name and w_file pointers to point to the string "(dead)" and the w_line field to 0. This way, if a mutex of a given name is used only in a module, then as long as all mutexes in the module are destroyed when the module is unloaded, witness will not maintain stale references to the mutex's name in the module's data section causing a panic later on when the w_name or w_file field's are examined.
2001-04-09 22:34:05 +00:00
w->w_refcount = 1;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
STAILQ_INSERT_HEAD(&w_all, w, w_list);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
if (lock_class->lc_flags & LC_SPINLOCK) {
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
STAILQ_INSERT_HEAD(&w_spin, w, w_typelist);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_spin_cnt++;
} else if (lock_class->lc_flags & LC_SLEEPLOCK) {
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
STAILQ_INSERT_HEAD(&w_sleep, w, w_typelist);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_sleep_cnt++;
Avoid endless recursion on panic. Reviewed by: jhb
2001-04-10 00:56:19 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* Insert new witness into the hash */
witness_hash_put(w);
witness_increment_graph_generation();
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_unlock_spin(&w_mtx);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
return (w);
found:
w->w_refcount++;
mtx_unlock_spin(&w_mtx);
if (lock_class != w->w_class)
panic(
"lock (%s) %s does not match earlier (%s) lock",
description, lock_class->lc_name,
w->w_class->lc_name);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (w);
}
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
static void
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
depart(struct witness *w)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
struct witness_list *list;
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
MPASS(w->w_refcount == 0);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
if (w->w_class->lc_flags & LC_SLEEPLOCK) {
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
list = &w_sleep;
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_sleep_cnt--;
} else {
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
list = &w_spin;
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_spin_cnt--;
}
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* Set file to NULL as it may point into a loadable module.
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
w->w_file = NULL;
w->w_line = 0;
witness_increment_graph_generation();
}
static void
adopt(struct witness *parent, struct witness *child)
{
int pi, ci, i, j;
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
/* If the relationship is already known, there's no work to be done. */
if (isitmychild(parent, child))
return;
/* When the structure of the graph changes, bump up the generation. */
witness_increment_graph_generation();
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* The hard part ... create the direct relationship, then propagate all
* indirect relationships.
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
pi = parent->w_index;
ci = child->w_index;
WITNESS_INDEX_ASSERT(pi);
WITNESS_INDEX_ASSERT(ci);
MPASS(pi != ci);
w_rmatrix[pi][ci] |= WITNESS_PARENT;
w_rmatrix[ci][pi] |= WITNESS_CHILD;
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
/*
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
* If parent was not already an ancestor of child,
* then we increment the descendant and ancestor counters.
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
*/
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if ((w_rmatrix[pi][ci] & WITNESS_ANCESTOR) == 0) {
parent->w_num_descendants++;
child->w_num_ancestors++;
}
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Find each ancestor of 'pi'. Note that 'pi' itself is counted as
* an ancestor of 'pi' during this loop.
*/
for (i = 1; i <= w_max_used_index; i++) {
if ((w_rmatrix[i][pi] & WITNESS_ANCESTOR_MASK) == 0 &&
(i != pi))
continue;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/* Find each descendant of 'i' and mark it as a descendant. */
for (j = 1; j <= w_max_used_index; j++) {
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Skip children that are already marked as
* descendants of 'i'.
*/
if (w_rmatrix[i][j] & WITNESS_ANCESTOR_MASK)
continue;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* We are only interested in descendants of 'ci'. Note
* that 'ci' itself is counted as a descendant of 'ci'.
*/
if ((w_rmatrix[ci][j] & WITNESS_ANCESTOR_MASK) == 0 &&
(j != ci))
continue;
w_rmatrix[i][j] |= WITNESS_ANCESTOR;
w_rmatrix[j][i] |= WITNESS_DESCENDANT;
w_data[i].w_num_descendants++;
w_data[j].w_num_ancestors++;
/*
* Make sure we aren't marking a node as both an
* ancestor and descendant. We should have caught
* this as a lock order reversal earlier.
*/
if ((w_rmatrix[i][j] & WITNESS_ANCESTOR_MASK) &&
(w_rmatrix[i][j] & WITNESS_DESCENDANT_MASK)) {
printf("witness rmatrix paradox! [%d][%d]=%d "
"both ancestor and descendant\n",
i, j, w_rmatrix[i][j]);
kdb_backtrace();
printf("Witness disabled.\n");
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = -1;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
}
if ((w_rmatrix[j][i] & WITNESS_ANCESTOR_MASK) &&
(w_rmatrix[j][i] & WITNESS_DESCENDANT_MASK)) {
printf("witness rmatrix paradox! [%d][%d]=%d "
"both ancestor and descendant\n",
j, i, w_rmatrix[j][i]);
kdb_backtrace();
printf("Witness disabled.\n");
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = -1;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
}
}
}
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
static void
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
itismychild(struct witness *parent, struct witness *child)
{
MPASS(child != NULL && parent != NULL);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
- Split the itismychild() function into two functions: insertchild() adds a witness to the child list of a parent witness. rebalancetree() runs through the entire tree removing direct descendants of witnesses who already have said child witness as an indirect descendant through another direct descendant. itismychild() now calls insertchild() followed by rebalancetree() and no longer needs the evil hack of having static recursed variable. - Add a function reparentchildren() that adds all the direct descendants of one witness as direct descendants of another witness. - Change the return value of itismychild() and similar functions so that they return 0 in the case of failure due to lack of resources instead of 1. This makes the return value more intuitive. - Check the return value of itismychild() when defining the static lock order in witness_initialize(). - Don't try to setup a lock instance in witness_lock() if itismychild() fails. Witness is hosed anyways so no need to do any more witness related activity at that point. It also makes the code flow easier to understand. - Add a new depart() function as the opposite of enroll(). When the reference count of a witness drops to 0 in witness_destroy(), this function is called on that witness. First, it runs through the lock order tree using reparentchildren() to reparent direct descendants of the departing witness to each of the witness' parents in the tree. Next, it releases it's own child list and other associated resources. Finally it calls rebalanacetree() to rebalance the lock order tree. - Sort function prototypes into something closer to alphabetical order. As a result of these changes, there should no longer be 'dead' witnesses in the order tree, and repeatedly loading and unloading a module should no longer exhaust witness of its internal resources. Inspired by: gallatin
2003-03-11 22:07:35 +00:00
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (!witness_lock_type_equal(parent, child)) {
if (witness_cold == 0)
mtx_unlock_spin(&w_mtx);
panic("%s: parent \"%s\" (%s) and child \"%s\" (%s) are not "
"the same lock type", __func__, parent->w_name,
parent->w_class->lc_name, child->w_name,
child->w_class->lc_name);
}
adopt(parent, child);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Generic code for the isitmy*() functions. The rmask parameter is the
* expected relationship of w1 to w2.
*/
static int
_isitmyx(struct witness *w1, struct witness *w2, int rmask, const char *fname)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
unsigned char r1, r2;
int i1, i2;
i1 = w1->w_index;
i2 = w2->w_index;
WITNESS_INDEX_ASSERT(i1);
WITNESS_INDEX_ASSERT(i2);
r1 = w_rmatrix[i1][i2] & WITNESS_RELATED_MASK;
r2 = w_rmatrix[i2][i1] & WITNESS_RELATED_MASK;
/* The flags on one better be the inverse of the flags on the other */
if (!((WITNESS_ATOD(r1) == r2 && WITNESS_DTOA(r2) == r1) ||
(WITNESS_DTOA(r1) == r2 && WITNESS_ATOD(r2) == r1))) {
printf("%s: rmatrix mismatch between %s (index %d) and %s "
"(index %d): w_rmatrix[%d][%d] == %hhx but "
"w_rmatrix[%d][%d] == %hhx\n",
fname, w1->w_name, i1, w2->w_name, i2, i1, i2, r1,
i2, i1, r2);
kdb_backtrace();
printf("Witness disabled.\n");
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = -1;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
}
return (r1 & rmask);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Checks if @child is a direct child of @parent.
*/
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
static int
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
isitmychild(struct witness *parent, struct witness *child)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
return (_isitmyx(parent, child, WITNESS_PARENT, __func__));
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
/*
* Checks if @descendant is a direct or inderect descendant of @ancestor.
*/
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
static int
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
isitmydescendant(struct witness *ancestor, struct witness *descendant)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
return (_isitmyx(ancestor, descendant, WITNESS_ANCESTOR_MASK,
__func__));
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#ifdef BLESSING
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
static int
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
blessed(struct witness *w1, struct witness *w2)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
int i;
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
struct witness_blessed *b;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
for (i = 0; i < blessed_count; i++) {
b = &blessed_list[i];
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (strcmp(w1->w_name, b->b_lock1) == 0) {
if (strcmp(w2->w_name, b->b_lock2) == 0)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (1);
continue;
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (strcmp(w1->w_name, b->b_lock2) == 0)
if (strcmp(w2->w_name, b->b_lock1) == 0)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (1);
}
return (0);
}
#unifdef the code for checking blessed lock collisions until we need it. Spotted by: DARPA & NAI Labs.
2002-10-20 08:48:39 +00:00
#endif
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
static struct witness *
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
witness_get(void)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
struct witness *w;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
int index;
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (witness_watch == -1) {
Check witness_dead in more functions to avoid panic'ing when assertions fail due to witness exhausting its internal resources and shutting down. Reported by: Szilveszter Adam <sziszi@petra.hos.u-szeged.hu> Tested by: David Wolfskill <david@catwhisker.org>
2001-05-11 20:25:29 +00:00
mtx_unlock_spin(&w_mtx);
return (NULL);
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
if (STAILQ_EMPTY(&w_free)) {
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = -1;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_unlock_spin(&w_mtx);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
printf("WITNESS: unable to allocate a new witness object\n");
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (NULL);
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
w = STAILQ_FIRST(&w_free);
STAILQ_REMOVE_HEAD(&w_free, w_list);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_free_cnt--;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
index = w->w_index;
MPASS(index > 0 && index == w_max_used_index+1 &&
index < WITNESS_COUNT);
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
bzero(w, sizeof(*w));
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
w->w_index = index;
if (index > w_max_used_index)
w_max_used_index = index;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
return (w);
}
static void
Remove the mtx_t, witness_t, and witness_blessed_t types. Instead, just use struct mtx, struct witness, and struct witness_blessed. Requested by: bde
2000-09-14 20:15:16 +00:00
witness_free(struct witness *w)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
STAILQ_INSERT_HEAD(&w_free, w, w_list);
Track all lock relationships instead of pruning direct relationships if an indirect relationship exists (keep both A->B->C and A->C). This allows witness_checkorder() to use isitmychild() instead of the much more expensive isitmydescendant() to check for valid lock ordering. Don't do an expensive tree walk to update the w_level values when the tree is updated. Only update the w_level values when using the debugger to display the tree. Nuke the experimental "witness_watch > 1" mode that only compared w_level for the two locks. This information is no longer maintained at run time, and the use of isitmychild() in witness_checkorder should bring performance close enough to the acceptable level that this hack is not needed. Report witness data structure allocation statistics under the debug.witness sysctl. Reviewed by: jhb MFC after: 30 days
2005-08-25 03:47:37 +00:00
w_free_cnt++;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
static struct lock_list_entry *
witness_lock_list_get(void)
{
struct lock_list_entry *lle;
Add a new ddb command 'witness_list' that lists the mutexes held by curproc. Requested by: peter
2001-01-27 07:51:34 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_watch == -1)
Check witness_dead in more functions to avoid panic'ing when assertions fail due to witness exhausting its internal resources and shutting down. Reported by: Szilveszter Adam <sziszi@petra.hos.u-szeged.hu> Tested by: David Wolfskill <david@catwhisker.org>
2001-05-11 20:25:29 +00:00
return (NULL);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_lock_spin(&w_mtx);
lle = w_lock_list_free;
if (lle == NULL) {
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = -1;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_unlock_spin(&w_mtx);
printf("%s: witness exhausted\n", __func__);
return (NULL);
}
w_lock_list_free = lle->ll_next;
mtx_unlock_spin(&w_mtx);
bzero(lle, sizeof(*lle));
return (lle);
}
static void
witness_lock_list_free(struct lock_list_entry *lle)
Add a new ddb command 'witness_list' that lists the mutexes held by curproc. Requested by: peter
2001-01-27 07:51:34 +00:00
{
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
mtx_lock_spin(&w_mtx);
lle->ll_next = w_lock_list_free;
w_lock_list_free = lle;
mtx_unlock_spin(&w_mtx);
Add a new ddb command 'witness_list' that lists the mutexes held by curproc. Requested by: peter
2001-01-27 07:51:34 +00:00
}
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
static struct lock_instance *
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
find_instance(struct lock_list_entry *list, const struct lock_object *lock)
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
{
struct lock_list_entry *lle;
struct lock_instance *instance;
int i;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
for (lle = list; lle != NULL; lle = lle->ll_next)
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
for (i = lle->ll_count - 1; i >= 0; i--) {
instance = &lle->ll_children[i];
if (instance->li_lock == lock)
return (instance);
}
return (NULL);
}
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
static void
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_lock(struct lock_instance *instance,
int (*prnt)(const char *fmt, ...))
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
{
struct lock_object *lock;
lock = instance->li_lock;
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
prnt("%s %s %s", (instance->li_flags & LI_EXCLUSIVE) != 0 ?
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
"exclusive" : "shared", LOCK_CLASS(lock)->lc_name, lock->lo_name);
- Embed the recursion counter for any locking primitive directly in the lock_object, using an unified field called lo_data. - Replace lo_type usage with the w_name usage and at init time pass the lock "type" directly to witness_init() from the parent lock init function. Handle delayed initialization before than witness_initialize() is called through the witness_pendhelp structure. - Axe out LO_ENROLLPEND as it is not really needed. The case where the mutex init delayed wants to be destroyed can't happen because witness_destroy() checks for witness_cold and panic in case. - In enroll(), if we cannot allocate a new object from the freelist, notify that to userspace through a printf(). - Modify the depart function in order to return nothing as in the current CVS version it always returns true and adjust callers accordingly. - Fix the witness_addgraph() argument name prototype. - Remove unuseful code from itismychild(). This commit leads to a shrinked struct lock_object and so smaller locks, in particular on amd64 where 2 uintptr_t (16 bytes per-primitive) are gained. Reviewed by: jhb
2008-05-15 20:10:06 +00:00
if (lock->lo_witness->w_name != lock->lo_name)
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
prnt(" (%s)", lock->lo_witness->w_name);
prnt(" r = %d (%p) locked @ %s:%d\n",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
instance->li_flags & LI_RECURSEMASK, lock,
- Fixup filenames in a few more places where they are used. - Some whitespace fixes.
2011-10-26 15:17:42 +00:00
fixup_filename(instance->li_file), instance->li_line);
A small overhaul of witness: - Add a comment about special lock order rules and Giant near the top of subr_witness.c. Specifically, this documents and explains the real lock order relationship between Giant and sleepable locks (i.e. lockmgr locks and sx locks). Basically, Giant can be safely acquired either before or after sleepable locks and the case of Giant before a sleepable lock is exempted as a special case. - Add a new static function 'witness_list_lock()' that displays a single line of information about a struct lock_instance. This is used to make the output of witness messages more consistent and reduce some code duplication. - Fixup a few comments in witness_lock(). - Properly handle the Giant-before-sleepable-lock lock order exception in a more general fashion and remove the no longer needed LI_SLEPT flag. - Break up the last condition before assuming a reversal a bit to try and make the logic less confusing in witness_lock(). - Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it with a more general WITNESS_WARN() macro/function combination. WITNESS_WARN() allows you to output a customized message out to the console along with a list of held locks. It will optionally drop into the debugger as well. You can exempt a single lock from the check by passing it in as the second argument. You can also use flags to specify if Giant should be exempt from the check, if all sleepable locks should be exempt from the check, and if witness should panic if any non-exempt locks are found. - Make the witness_list() function static. Other areas of the kernel should use the new WITNESS_WARN() instead.
2003-03-04 20:56:39 +00:00
}
When DDB is not defined, don't implement witness_thread_has_locks() and witness_proc_has_locks(), as they are unused, which results in a compiler error. This problem was introduced with the implementation of "show alllocks". Spotted by: Artem Kuchin <matrix at itlegion dot ru>
2005-01-22 21:14:21 +00:00
#ifdef DDB
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
static int
witness_thread_has_locks(struct thread *td)
{
- For any lock list we hold the head in order to reduce allocation from the free list and in this way avoid contention on the w_mtx. In order to make the code simple, we rely on the rule that when the head has not a child it also doesn't have other subsequent entries. Actually this assertion is broken because we can free all the head children and quit witness_unlock() with the head still allocated, with no children and subsequent entries present. Fix this by shifting the head if other entries are present and still freeing the object, but leaving always an head. - Fix witness_thread_has_locks() in order to report, correctly, if the lock list linked to a specific thread has children or not based on the above explained rule. - Fix a printout into DDB's "show alllocks" command in order to show, correctly, the process name that is really what we want. - Fix style(9) for a comment. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> Reported by: Marko Kiiskila <marko dot kiiskila at nokia dot com> Sponsored by: Nokia
2008-09-12 21:44:01 +00:00
if (td->td_sleeplocks == NULL)
return (0);
return (td->td_sleeplocks->ll_count != 0);
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
}
static int
witness_proc_has_locks(struct proc *p)
{
struct thread *td;
FOREACH_THREAD_IN_PROC(p, td) {
if (witness_thread_has_locks(td))
return (1);
}
return (0);
}
When DDB is not defined, don't implement witness_thread_has_locks() and witness_proc_has_locks(), as they are unused, which results in a compiler error. This problem was introduced with the implementation of "show alllocks". Spotted by: Artem Kuchin <matrix at itlegion dot ru>
2005-01-22 21:14:21 +00:00
#endif
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
int
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_locks(struct lock_list_entry **lock_list,
int (*prnt)(const char *fmt, ...))
- Change the 'witness_list' ddb command to 'show mutexes'. Note that this will only display sleep mutexes held by the current process. - Clean up some nits in the witness_display() function and add a ddb command 'show witness' that dumps the hierarchy and order lists to the console. - Use queue(3) macros where appropriate. - Resort the spin lock order list so that "com" is before "sched_lock". Also, add appropriate #ifdef's around SMP and i386-specific mutexes. - Add two new mutexes used to protect the ithread lists and tables to the order list. Requested by: bde (1)
2001-02-09 15:19:41 +00:00
{
- Split out the functionality of displaying the contents of a single lock list into a public witness_list_locks() function. Call this function twice in witness_list() instead of using an evil goto. - Adjust the 'show locks' command to take an optional parameter which specifies the pid of a process to list the locks of. By default the locks held by the current process are displayed.
2001-04-06 21:37:52 +00:00
struct lock_list_entry *lle;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
int i, nheld;
nheld = 0;
for (lle = *lock_list; lle != NULL; lle = lle->ll_next)
for (i = lle->ll_count - 1; i >= 0; i--) {
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_lock(&lle->ll_children[i], prnt);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
nheld++;
}
- Split out the functionality of displaying the contents of a single lock list into a public witness_list_locks() function. Call this function twice in witness_list() instead of using an evil goto. - Adjust the 'show locks' command to take an optional parameter which specifies the pid of a process to list the locks of. By default the locks held by the current process are displayed.
2001-04-06 21:37:52 +00:00
return (nheld);
}
Add a new function to look for a spinlock's instance when it is held by another thread. We use the td_oncpu member of the other field to locate it's associated CPU and then search the that CPU's list of spin locks contained in its per-CPU data. This is not always safe and may in fact panic or just not work, but it is useful in at least one case.
2003-07-31 18:50:58 +00:00
/*
* This is a bit risky at best. We call this function when we have timed
* out acquiring a spin lock, and we assume that the other CPU is stuck
* with this lock held. So, we go groveling around in the other CPU's
* per-cpu data to try to find the lock instance for this spin lock to
* see when it was last acquired.
*/
void
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_display_spinlock(struct lock_object *lock, struct thread *owner,
int (*prnt)(const char *fmt, ...))
Add a new function to look for a spinlock's instance when it is held by another thread. We use the td_oncpu member of the other field to locate it's associated CPU and then search the that CPU's list of spin locks contained in its per-CPU data. This is not always safe and may in fact panic or just not work, but it is useful in at least one case.
2003-07-31 18:50:58 +00:00
{
struct lock_instance *instance;
struct pcpu *pc;
if (owner->td_critnest == 0 || owner->td_oncpu == NOCPU)
return;
pc = pcpu_find(owner->td_oncpu);
instance = find_instance(pc->pc_spinlocks, lock);
if (instance != NULL)
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_lock(instance, prnt);
Add a new function to look for a spinlock's instance when it is held by another thread. We use the td_oncpu member of the other field to locate it's associated CPU and then search the that CPU's list of spin locks contained in its per-CPU data. This is not always safe and may in fact panic or just not work, but it is useful in at least one case.
2003-07-31 18:50:58 +00:00
}
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
void
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
witness_save(struct lock_object *lock, const char **filep, int *linep)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Teach WITNESS_SAVE() and WITNESS_RESTORE() to work with spin locks instead of only sleep locks.
2005-12-29 20:54:25 +00:00
struct lock_list_entry *lock_list;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
struct lock_instance *instance;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
struct lock_class *class;
Remove MUTEX_DECLARE() and MTX_COLD. Instead, postpone full mutex initialization until after malloc() is safe to call, then iterate through all mutexes and complete their initialization. This change is necessary in order to avoid some circular bootstrapping dependencies.
2001-01-21 07:52:20 +00:00
panic: add a switch and infrastructure for stopping other CPUs in SMP case Historical behavior of letting other CPUs merily go on is a default for time being. The new behavior can be switched on via kern.stop_scheduler_on_panic tunable and sysctl. Stopping of the CPUs has (at least) the following benefits: - more of the system state at panic time is preserved intact - threads and interrupts do not interfere with dumping of the system state Only one thread runs uninterrupted after panic if stop_scheduler_on_panic is set. That thread might call code that is also used in normal context and that code might use locks to prevent concurrent execution of certain parts. Those locks might be held by the stopped threads and would never be released. To work around this issue, it was decided that instead of explicit checks for panic context, we would rather put those checks inside the locking primitives. This change has substantial portions written and re-written by attilio and kib at various times. Other changes are heavily based on the ideas and patches submitted by jhb and mdf. bde has provided many insights into the details and history of the current code. The new behavior may cause problems for systems that use a USB keyboard for interfacing with system console. This is because of some unusual locking patterns in the ukbd code which have to be used because on one hand ukbd is below syscons, but on the other hand it has to interface with other usb code that uses regular mutexes/Giant for its concurrency protection. Dumping to USB-connected disks may also be affected. PR: amd64/139614 (at least) In cooperation with: attilio, jhb, kib, mdf Discussed with: arch@, bde Tested by: Eugene Grosbein <eugen@grosbein.net>, gnn, Steven Hartland <killing@multiplay.co.uk>, glebius, Andrew Boyer <aboyer@averesystems.com> (various versions of the patch) MFC after: 3 months (or never)
2011-12-11 21:02:01 +00:00
/*
* This function is used independently in locking code to deal with
* Giant, SCHEDULER_STOPPED() check can be removed here after Giant
* is gone.
*/
if (SCHEDULER_STOPPED())
return;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (lock->lo_witness == NULL || witness_watch == -1 || panicstr != NULL)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
return;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
if (class->lc_flags & LC_SLEEPLOCK)
Teach WITNESS_SAVE() and WITNESS_RESTORE() to work with spin locks instead of only sleep locks.
2005-12-29 20:54:25 +00:00
lock_list = curthread->td_sleeplocks;
else {
if (witness_skipspin)
return;
lock_list = PCPU_GET(spinlocks);
}
instance = find_instance(lock_list, lock);
Convert some KASSERT()'s into if (foo) panic() because they are testing how locks are managed by the rest of the kernel, not verifying the internal integrity of witness itself.
2001-08-23 22:44:47 +00:00
if (instance == NULL)
panic("%s: lock (%s) %s not locked", __func__,
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class->lc_name, lock->lo_name);
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
*filep = instance->li_file;
*linep = instance->li_line;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
void
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
witness_restore(struct lock_object *lock, const char *file, int line)
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
{
Teach WITNESS_SAVE() and WITNESS_RESTORE() to work with spin locks instead of only sleep locks.
2005-12-29 20:54:25 +00:00
struct lock_list_entry *lock_list;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
struct lock_instance *instance;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
struct lock_class *class;
Remove MUTEX_DECLARE() and MTX_COLD. Instead, postpone full mutex initialization until after malloc() is safe to call, then iterate through all mutexes and complete their initialization. This change is necessary in order to avoid some circular bootstrapping dependencies.
2001-01-21 07:52:20 +00:00
panic: add a switch and infrastructure for stopping other CPUs in SMP case Historical behavior of letting other CPUs merily go on is a default for time being. The new behavior can be switched on via kern.stop_scheduler_on_panic tunable and sysctl. Stopping of the CPUs has (at least) the following benefits: - more of the system state at panic time is preserved intact - threads and interrupts do not interfere with dumping of the system state Only one thread runs uninterrupted after panic if stop_scheduler_on_panic is set. That thread might call code that is also used in normal context and that code might use locks to prevent concurrent execution of certain parts. Those locks might be held by the stopped threads and would never be released. To work around this issue, it was decided that instead of explicit checks for panic context, we would rather put those checks inside the locking primitives. This change has substantial portions written and re-written by attilio and kib at various times. Other changes are heavily based on the ideas and patches submitted by jhb and mdf. bde has provided many insights into the details and history of the current code. The new behavior may cause problems for systems that use a USB keyboard for interfacing with system console. This is because of some unusual locking patterns in the ukbd code which have to be used because on one hand ukbd is below syscons, but on the other hand it has to interface with other usb code that uses regular mutexes/Giant for its concurrency protection. Dumping to USB-connected disks may also be affected. PR: amd64/139614 (at least) In cooperation with: attilio, jhb, kib, mdf Discussed with: arch@, bde Tested by: Eugene Grosbein <eugen@grosbein.net>, gnn, Steven Hartland <killing@multiplay.co.uk>, glebius, Andrew Boyer <aboyer@averesystems.com> (various versions of the patch) MFC after: 3 months (or never)
2011-12-11 21:02:01 +00:00
/*
* This function is used independently in locking code to deal with
* Giant, SCHEDULER_STOPPED() check can be removed here after Giant
* is gone.
*/
if (SCHEDULER_STOPPED())
return;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
if (lock->lo_witness == NULL || witness_watch == -1 || panicstr != NULL)
Move most of sys/mutex.h into kern/kern_mutex.c, thereby making the mutex inline functions non-inlined. Hide parts of the mutex implementation that should not be exposed. Make sure that WITNESS code is not executed during boot until the mutexes are fully initialized by SI_SUB_MUTEX (the original motivation for this commit). Submitted by: peter
2001-01-21 22:34:43 +00:00
return;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
if (class->lc_flags & LC_SLEEPLOCK)
Teach WITNESS_SAVE() and WITNESS_RESTORE() to work with spin locks instead of only sleep locks.
2005-12-29 20:54:25 +00:00
lock_list = curthread->td_sleeplocks;
else {
if (witness_skipspin)
return;
lock_list = PCPU_GET(spinlocks);
}
instance = find_instance(lock_list, lock);
Convert some KASSERT()'s into if (foo) panic() because they are testing how locks are managed by the rest of the kernel, not verifying the internal integrity of witness itself.
2001-08-23 22:44:47 +00:00
if (instance == NULL)
panic("%s: lock (%s) %s not locked", __func__,
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class->lc_name, lock->lo_name);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
lock->lo_witness->w_file = file;
lock->lo_witness->w_line = line;
- Move state about lock objects out of struct lock_object and into a new struct lock_instance that is stored in the per-process and per-CPU lock lists. Previously, the lock lists just kept a pointer to each lock held. That pointer is now replaced by a lock instance which contains a pointer to the lock object, the file and line of the last acquisition of a lock, and various flags about a lock including its recursion count. - If we sleep while holding a sleepable lock, then mark that lock instance as having slept and ignore any lock order violations that occur while acquiring Giant when we wake up with slept locks. This is ok because of Giant's special nature. - Allow witness to differentiate between shared and exclusive locks and unlocks of a lock. Witness will now detect the case when a lock is acquired first in one mode and then in another. Mutexes are always locked and unlocked exclusively. Witness will also now detect the case where a process attempts to unlock a shared lock while holding an exclusive lock and vice versa. - Fix a bug in the lock list implementation where we used the wrong constant to detect the case where a lock list entry was full.
2001-05-04 17:15:16 +00:00
instance->li_file = file;
instance->li_line = line;
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
void
Constify arguments for locking KPIs where possible. This enables locking consumers to pass their own structures around as const and be able to assert locks embedded into those structures. Reviewed by: ed, kib, jhb
2011-11-16 21:51:17 +00:00
witness_assert(const struct lock_object *lock, int flags, const char *file,
int line)
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
{
#ifdef INVARIANT_SUPPORT
struct lock_instance *instance;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
struct lock_class *class;
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (lock->lo_witness == NULL || witness_watch < 1 || panicstr != NULL)
Don't check witness assertions if the lock doesn't use witness or witness is dead.
2001-06-28 22:22:20 +00:00
return;
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class = LOCK_CLASS(lock);
if ((class->lc_flags & LC_SLEEPLOCK) != 0)
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
instance = find_instance(curthread->td_sleeplocks, lock);
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
else if ((class->lc_flags & LC_SPINLOCK) != 0)
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
instance = find_instance(PCPU_GET(spinlocks), lock);
Add a couple of returns to making recovering from a failed witness_assert() more sane in the RESTARTABLE_PANICS case.
2001-11-15 19:46:36 +00:00
else {
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
panic("Lock (%s) %s is not sleep or spin!",
Trim another pointer from struct lock_object (and thus from struct mtx and struct sx). Instead of storing a direct pointer to a our lock_class struct in lock_object, reserve 4 bits in the lo_flags field to serve as an index into a global lock_classes array that contains pointers to the lock classes. Only debugging code such as WITNESS or INVARIANTS checks and KTR logging need to access the lock_class member, so this shouldn't add any overhead to production kernels. It might add some slight overhead to kernels using those debug options however. As with the previous set of changes to lock_object, this is going to completely obliterate the kernel ABI, so be sure to recompile all your modules.
2006-01-06 18:07:32 +00:00
class->lc_name, lock->lo_name);
Add a couple of returns to making recovering from a failed witness_assert() more sane in the RESTARTABLE_PANICS case.
2001-11-15 19:46:36 +00:00
}
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
switch (flags) {
case LA_UNLOCKED:
if (instance != NULL)
panic("Lock (%s) %s locked @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
break;
case LA_LOCKED:
case LA_LOCKED | LA_RECURSED:
case LA_LOCKED | LA_NOTRECURSED:
case LA_SLOCKED:
case LA_SLOCKED | LA_RECURSED:
case LA_SLOCKED | LA_NOTRECURSED:
case LA_XLOCKED:
case LA_XLOCKED | LA_RECURSED:
case LA_XLOCKED | LA_NOTRECURSED:
Add a couple of returns to making recovering from a failed witness_assert() more sane in the RESTARTABLE_PANICS case.
2001-11-15 19:46:36 +00:00
if (instance == NULL) {
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
panic("Lock (%s) %s not locked @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
Add a couple of returns to making recovering from a failed witness_assert() more sane in the RESTARTABLE_PANICS case.
2001-11-15 19:46:36 +00:00
break;
}
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
if ((flags & LA_XLOCKED) != 0 &&
(instance->li_flags & LI_EXCLUSIVE) == 0)
panic("Lock (%s) %s not exclusively locked @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
if ((flags & LA_SLOCKED) != 0 &&
(instance->li_flags & LI_EXCLUSIVE) != 0)
panic("Lock (%s) %s exclusively locked @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
if ((flags & LA_RECURSED) != 0 &&
(instance->li_flags & LI_RECURSEMASK) == 0)
panic("Lock (%s) %s not recursed @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
if ((flags & LA_NOTRECURSED) != 0 &&
(instance->li_flags & LI_RECURSEMASK) != 0)
panic("Lock (%s) %s recursed @ %s:%d.",
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
class->lc_name, lock->lo_name,
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
break;
default:
Don't call fixup_filename() on each witness lock call. This has been irking me for a while. This causes significant CPU use on bottlenecked CPUs (eg my older EEEPC w/ an earlier Celeron CPU and my MIPS24k boards) when they're passing a lot of traffic. Since the file/line values are only used for printing, this should only affect display. It should have no operational change on the code, besides reducing CPU use.
2011-10-12 09:21:02 +00:00
panic("Invalid lock assertion at %s:%d.",
fixup_filename(file), line);
- Add a new witness_assert() to perform arbitrary locking assertions. - Clean up the KTR tracepoints to be slighlty more consistent and useful - Fix a bug in WITNESS where we would recurse indefinitely and blow the stack when acquiring Giant after sleeping with a sleepable lock held. Reported by: tanimura (3)
2001-06-27 06:27:29 +00:00
}
#endif /* INVARIANT_SUPPORT */
}
Add functions WITNESS so it can be asserted that the lock is not released for a section of code, this uses WITNESS_NORELEASE() and WITNESS_RELEASEOK() to mark the boundaries. Both functions require the lock to be held when calling. This is intended for scenarios like a bus asserting that the bus lock is not dropped during a driver call. There doesn't appear to be a man page to document this in. Reviewed by: jhb
2009-01-21 04:19:18 +00:00
static void
witness_setflag(struct lock_object *lock, int flag, int set)
{
struct lock_list_entry *lock_list;
struct lock_instance *instance;
struct lock_class *class;
if (lock->lo_witness == NULL || witness_watch == -1 || panicstr != NULL)
return;
class = LOCK_CLASS(lock);
if (class->lc_flags & LC_SLEEPLOCK)
lock_list = curthread->td_sleeplocks;
else {
if (witness_skipspin)
return;
lock_list = PCPU_GET(spinlocks);
}
instance = find_instance(lock_list, lock);
if (instance == NULL)
panic("%s: lock (%s) %s not locked", __func__,
class->lc_name, lock->lo_name);
if (set)
instance->li_flags |= flag;
else
instance->li_flags &= ~flag;
}
void
witness_norelease(struct lock_object *lock)
{
witness_setflag(lock, LI_NORELEASE, 1);
}
void
witness_releaseok(struct lock_object *lock)
{
witness_setflag(lock, LI_NORELEASE, 0);
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#ifdef DDB
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
static void
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_list(struct thread *td)
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
KASSERT(witness_cold == 0, ("%s: witness_cold", __func__));
Update for the KDB framework: o Make debugging code conditional upon KDB instead of DDB. o s/WITNESS_DDB/WITNESS_KDB/g o s/witness_ddb/witness_kdb/g o Rename the debug.witness_ddb sysctl to debug.witness_kdb. o Call kdb_backtrace() instead of backtrace(). o Call kdb_enter() instead Debugger(). o Assert kdb_active instead of db_active.
2004-07-10 21:42:16 +00:00
KASSERT(kdb_active, ("%s: not in the debugger", __func__));
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_watch < 1)
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
return;
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_locks(&td->td_sleeplocks, db_printf);
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
/*
* We only handle spinlocks if td == curthread. This is somewhat broken
* if td is currently executing on some other CPU and holds spin locks
* as we won't display those locks. If we had a MI way of getting
* the per-cpu data for a given cpu then we could use
Move the _oncpu entry from the KSE to the thread. The entry in the KSE still exists but it's purpose will change a bit when we add the ability to lock a KSE to a cpu.
2003-04-10 17:35:44 +00:00
* td->td_oncpu to get the list of spinlocks for this thread
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
* and "fix" this.
*
Commit 10/14 of sched_lock decomposition. - Add new spinlocks to support thread_lock() and adjust ordering. Tested by: kris, current@ Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc. Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
2007-06-04 23:55:45 +00:00
* That still wouldn't really fix this unless we locked the scheduler
* lock or stopped the other CPU to make sure it wasn't changing the
* list out from under us. It is probably best to just not try to
* handle threads on other CPU's for now.
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
*/
if (td == curthread && PCPU_GET(spinlocks) != NULL)
Right now, WITNESS just blindly pipes all the output to the (TOCONS | TOLOG) mask even when called from DDB points. That breaks several output, where the most notable is textdump output. Fix this by having configurable callbacks passed to witness_list_locks() and witness_display_spinlock() for printing out datas. Reported by: several broken textdump outputs Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> MFC after: 7 days X-MFC: r207922
2010-05-11 18:24:22 +00:00
witness_list_locks(PCPU_PTR(spinlocks), db_printf);
Now that we have WITNESS_WARN(), we only call witness_list() from the ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB section and remove extra checks for calling this function outside of DDB. Also, witness_list() now returns void instead of returning an int. Reported by: Steve Ames <steve@energistic.com> Prodded by: davidxu
2003-03-10 17:03:57 +00:00
}
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
- s/mutexes/locks/g in appropriate comments. - Rename the 'show mutexes' ddb command to 'show locks' since it shows a list of all the lock objects held by the current process.
2001-03-28 12:39:40 +00:00
DB_SHOW_COMMAND(locks, db_witness_list)
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
{
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td;
- Split out the functionality of displaying the contents of a single lock list into a public witness_list_locks() function. Call this function twice in witness_list() instead of using an evil goto. - Adjust the 'show locks' command to take an optional parameter which specifies the pid of a process to list the locks of. By default the locks held by the current process are displayed.
2001-04-06 21:37:52 +00:00
Use db_lookup_thread() to lookup the thread for the passed in address and change 'show locks' to only list the locks for a given thread rather than for all the threads in the process containing a specified thread.
2006-04-25 20:24:23 +00:00
if (have_addr)
td = db_lookup_thread(addr, TRUE);
else
td = kdb_thread;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_list(td);
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
}
Make ddb command registration dynamic so modules can extend the command set (only so long as the module is present): o add db_command_register and db_command_unregister to add and remove commands, respectively o replace linker sets with SYSINIT's (and SYSUINIT's) that register commands o expose 3 list heads: db_cmd_table, db_show_table, and db_show_all_table for registering top-level commands, show operands, and show all operands, respectively While here also: o sort command lists o add DB_ALIAS, DB_SHOW_ALIAS, and DB_SHOW_ALL_ALIAS to add aliases for existing commands o add "show all trace" as an alias for "show alltrace" o add "show all locks" as an alias for "show alllocks" Submitted by: Guillaume Ballet <gballet@gmail.com> (original version) Reviewed by: jhb MFC after: 1 month
2008-09-15 22:45:14 +00:00
DB_SHOW_ALL_COMMAND(locks, db_witness_list_all)
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
{
struct thread *td;
struct proc *p;
/*
* It would be nice to list only threads and processes that actually
* held sleep locks, but that information is currently not exported
* by WITNESS.
*/
FOREACH_PROC_IN_SYSTEM(p) {
if (!witness_proc_has_locks(p))
continue;
FOREACH_THREAD_IN_PROC(p, td) {
if (!witness_thread_has_locks(td))
continue;
Fix a deadlock I introduced with the recently added printf to warn about spin locks that are not in the static order list. It is not safe to call printf while holding the witness spin mutex since the console drivers that back printf may need to use their own spin locks which would try to talk to witness when they were locked. Given this, it is possible for one CPU to lock a console driver lock (such as sio) which then tries to lock the witness lock while another CPU is doing the printf while holding the witness lock. Fix this by moving the printf outside of the witness lock. All other printf's in witness are already correct. MFC after: 3 days
2005-12-29 20:53:01 +00:00
db_printf("Process %d (%s) thread %p (%d)\n", p->p_pid,
- For any lock list we hold the head in order to reduce allocation from the free list and in this way avoid contention on the w_mtx. In order to make the code simple, we rely on the rule that when the head has not a child it also doesn't have other subsequent entries. Actually this assertion is broken because we can free all the head children and quit witness_unlock() with the head still allocated, with no children and subsequent entries present. Fix this by shifting the head if other entries are present and still freeing the object, but leaving always an head. - Fix witness_thread_has_locks() in order to report, correctly, if the lock list linked to a specific thread has children or not based on the above explained rule. - Fix a printout into DDB's "show alllocks" command in order to show, correctly, the process name that is really what we want. - Fix style(9) for a comment. Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com> Reported by: Marko Kiiskila <marko dot kiiskila at nokia dot com> Sponsored by: Nokia
2008-09-12 21:44:01 +00:00
p->p_comm, td, td->td_tid);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_list(td);
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
}
}
}
Make ddb command registration dynamic so modules can extend the command set (only so long as the module is present): o add db_command_register and db_command_unregister to add and remove commands, respectively o replace linker sets with SYSINIT's (and SYSUINIT's) that register commands o expose 3 list heads: db_cmd_table, db_show_table, and db_show_all_table for registering top-level commands, show operands, and show all operands, respectively While here also: o sort command lists o add DB_ALIAS, DB_SHOW_ALIAS, and DB_SHOW_ALL_ALIAS to add aliases for existing commands o add "show all trace" as an alias for "show alltrace" o add "show all locks" as an alias for "show alllocks" Submitted by: Guillaume Ballet <gballet@gmail.com> (original version) Reviewed by: jhb MFC after: 1 month
2008-09-15 22:45:14 +00:00
DB_SHOW_ALIAS(alllocks, db_witness_list_all)
Add "show alllocks" command to DDB, which dumps a list of processes and threads currently holding sleep mutexes (and spin mutexes for curthread). This can be quite useful in looking for a lock condition summary for a system, as it avoids manually iterating through threads and processes to find all the interesting locks. NB: "alllocks" is up there with "lockedvnods" for a bad argument for show. MFC after: 2 weeks
2004-12-26 22:52:24 +00:00
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
DB_SHOW_COMMAND(witness, db_witness_display)
{
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
witness_ddb_display(db_printf);
}
#endif
static int
sysctl_debug_witness_badstacks(SYSCTL_HANDLER_ARGS)
{
struct witness_lock_order_data *data1, *data2, *tmp_data1, *tmp_data2;
struct witness *tmp_w1, *tmp_w2, *w1, *w2;
struct sbuf *sb;
u_int w_rmatrix1, w_rmatrix2;
int error, generation, i, j;
tmp_data1 = NULL;
tmp_data2 = NULL;
tmp_w1 = NULL;
tmp_w2 = NULL;
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_watch < 1) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
error = SYSCTL_OUT(req, w_notrunning, sizeof(w_notrunning));
return (error);
}
if (witness_cold) {
error = SYSCTL_OUT(req, w_stillcold, sizeof(w_stillcold));
return (error);
}
error = 0;
sb = sbuf_new(NULL, NULL, BADSTACK_SBUF_SIZE, SBUF_AUTOEXTEND);
if (sb == NULL)
return (ENOMEM);
/* Allocate and init temporary storage space. */
tmp_w1 = malloc(sizeof(struct witness), M_TEMP, M_WAITOK | M_ZERO);
tmp_w2 = malloc(sizeof(struct witness), M_TEMP, M_WAITOK | M_ZERO);
tmp_data1 = malloc(sizeof(struct witness_lock_order_data), M_TEMP,
M_WAITOK | M_ZERO);
tmp_data2 = malloc(sizeof(struct witness_lock_order_data), M_TEMP,
M_WAITOK | M_ZERO);
stack_zero(&tmp_data1->wlod_stack);
stack_zero(&tmp_data2->wlod_stack);
restart:
mtx_lock_spin(&w_mtx);
generation = w_generation;
mtx_unlock_spin(&w_mtx);
sbuf_printf(sb, "Number of known direct relationships is %d\n",
w_lohash.wloh_count);
for (i = 1; i < w_max_used_index; i++) {
mtx_lock_spin(&w_mtx);
if (generation != w_generation) {
mtx_unlock_spin(&w_mtx);
/* The graph has changed, try again. */
req->oldidx = 0;
sbuf_clear(sb);
goto restart;
}
w1 = &w_data[i];
if (w1->w_reversed == 0) {
mtx_unlock_spin(&w_mtx);
continue;
}
/* Copy w1 locally so we can release the spin lock. */
*tmp_w1 = *w1;
mtx_unlock_spin(&w_mtx);
if (tmp_w1->w_reversed == 0)
continue;
for (j = 1; j < w_max_used_index; j++) {
if ((w_rmatrix[i][j] & WITNESS_REVERSAL) == 0 || i > j)
continue;
mtx_lock_spin(&w_mtx);
if (generation != w_generation) {
mtx_unlock_spin(&w_mtx);
/* The graph has changed, try again. */
req->oldidx = 0;
sbuf_clear(sb);
goto restart;
}
w2 = &w_data[j];
data1 = witness_lock_order_get(w1, w2);
data2 = witness_lock_order_get(w2, w1);
/*
* Copy information locally so we can release the
* spin lock.
*/
*tmp_w2 = *w2;
w_rmatrix1 = (unsigned int)w_rmatrix[i][j];
w_rmatrix2 = (unsigned int)w_rmatrix[j][i];
if (data1) {
stack_zero(&tmp_data1->wlod_stack);
stack_copy(&data1->wlod_stack,
&tmp_data1->wlod_stack);
}
if (data2 && data2 != data1) {
stack_zero(&tmp_data2->wlod_stack);
stack_copy(&data2->wlod_stack,
&tmp_data2->wlod_stack);
}
mtx_unlock_spin(&w_mtx);
sbuf_printf(sb,
"\nLock order reversal between \"%s\"(%s) and \"%s\"(%s)!\n",
tmp_w1->w_name, tmp_w1->w_class->lc_name,
tmp_w2->w_name, tmp_w2->w_class->lc_name);
#if 0
sbuf_printf(sb,
"w_rmatrix[%s][%s] == %x, w_rmatrix[%s][%s] == %x\n",
tmp_w1->name, tmp_w2->w_name, w_rmatrix1,
tmp_w2->name, tmp_w1->w_name, w_rmatrix2);
#endif
if (data1) {
sbuf_printf(sb,
"Lock order \"%s\"(%s) -> \"%s\"(%s) first seen at:\n",
tmp_w1->w_name, tmp_w1->w_class->lc_name,
tmp_w2->w_name, tmp_w2->w_class->lc_name);
stack_sbuf_print(sb, &tmp_data1->wlod_stack);
sbuf_printf(sb, "\n");
}
if (data2 && data2 != data1) {
sbuf_printf(sb,
"Lock order \"%s\"(%s) -> \"%s\"(%s) first seen at:\n",
tmp_w2->w_name, tmp_w2->w_class->lc_name,
tmp_w1->w_name, tmp_w1->w_class->lc_name);
stack_sbuf_print(sb, &tmp_data2->wlod_stack);
sbuf_printf(sb, "\n");
}
}
}
mtx_lock_spin(&w_mtx);
if (generation != w_generation) {
mtx_unlock_spin(&w_mtx);
/*
* The graph changed while we were printing stack data,
* try again.
*/
req->oldidx = 0;
sbuf_clear(sb);
goto restart;
}
mtx_unlock_spin(&w_mtx);
/* Free temporary storage space. */
free(tmp_data1, M_TEMP);
free(tmp_data2, M_TEMP);
free(tmp_w1, M_TEMP);
free(tmp_w2, M_TEMP);
sbuf_finish(sb);
error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
sbuf_delete(sb);
return (error);
}
static int
sysctl_debug_witness_fullgraph(SYSCTL_HANDLER_ARGS)
{
struct witness *w;
struct sbuf *sb;
int error;
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (witness_watch < 1) {
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
error = SYSCTL_OUT(req, w_notrunning, sizeof(w_notrunning));
return (error);
}
if (witness_cold) {
error = SYSCTL_OUT(req, w_stillcold, sizeof(w_stillcold));
return (error);
}
error = 0;
Explicitly wire the user buffer rather than doing it implicitly in sbuf_new_for_sysctl(9). This allows using an sbuf with a SYSCTL_OUT drain for extremely large amounts of data where the caller knows that appropriate references are held, and sleeping is not an issue. Inspired by: rwatson
2011-01-27 00:34:12 +00:00
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
Re-add r212370 now that the LOR in powerpc64 has been resolved: Add a drain function for struct sysctl_req, and use it for a variety of handlers, some of which had to do awkward things to get a large enough SBUF_FIXEDLEN buffer. Note that some sysctl handlers were explicitly outputting a trailing NUL byte. This behaviour was preserved, though it should not be necessary. Reviewed by: phk (original patch)
2010-09-16 16:13:12 +00:00
sb = sbuf_new_for_sysctl(NULL, NULL, FULLGRAPH_SBUF_SIZE, req);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
if (sb == NULL)
return (ENOMEM);
sbuf_printf(sb, "\n");
mtx_lock_spin(&w_mtx);
STAILQ_FOREACH(w, &w_all, w_list)
w->w_displayed = 0;
STAILQ_FOREACH(w, &w_all, w_list)
witness_add_fullgraph(sb, w);
mtx_unlock_spin(&w_mtx);
/*
* Close the sbuf and return to userland.
*/
Re-add r212370 now that the LOR in powerpc64 has been resolved: Add a drain function for struct sysctl_req, and use it for a variety of handlers, some of which had to do awkward things to get a large enough SBUF_FIXEDLEN buffer. Note that some sysctl handlers were explicitly outputting a trailing NUL byte. This behaviour was preserved, though it should not be necessary. Reviewed by: phk (original patch)
2010-09-16 16:13:12 +00:00
error = sbuf_finish(sb);
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
sbuf_delete(sb);
return (error);
}
static int
sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS)
{
int error, value;
- Improve some witness_watch operability in code which does perform both lock tracking and checks, doing just the former ones. - Fix a bug where sysctl utility was printing crazy values when setting a new value for debug.witness.watch [0] [0] Reported by: yongari
2008-08-30 13:20:35 +00:00
value = witness_watch;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
error = sysctl_handle_int(oidp, &value, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
if (value > 1 || value < -1 ||
(witness_watch == -1 && value != witness_watch))
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
return (EINVAL);
- Make witness_watch a 3 state value. 1 means that witness is up and running. 0 means that witness is disabled but that it can be established later again in effective way. -1 means that witness is disabled permanently - Fix a bug causing kernel to panic on witness disabling through witness_watch. lock lists queues were still full of entries and this was causing throubles with debugging stubs (like witness_thread_exit()). Reported by: kris, yongari Sponsored by: Nokia
2008-08-29 15:47:53 +00:00
witness_watch = value;
Introduce some WITNESS improvements: - Speedup the lock orderings lookup modifying the witness graph from a linked tree to a matrix. A table lookup caches the lock orderings in order to make a O(1) access for them. Any witness object has an unique index withing this lookup cache table. - Reduce the lock contention on w_mtx acquiring it only when the LOR actually happens and not in a sane case. In order to do this don't totally flush lock lists (per-CPU spinlocks list and per-thread sleeplocks list) but check for ll_count anytime we need to have to verify allocations sanity. - Introduce the function witness_thread_exit() in the witness namespace which should verify a thread doesn't hold any witness occurrence why exiting. - Rename the sysctl debug.witness.graphs into debug.witness.fullgraph and add debug.witness.badstacks which prints out stacks for LOR revealed. This is implemented using the stack(9) support, which makes WITNESS to be dependent by the STACK option or by the DDB (including STACK) option. - Fix style(9) for src/sys/kern/subr_witness.c The hash table approach has been developed by Ilya Maykov on the behalf of Isilon Systems which kindly released the patch. Jeff Roberson, ported the patch to -CURRENT and fixed w_mtx contention, on the behalf of Nokia. Submitted by: Ilya Maykov <ivmaykov at gmail dot com> (Isilon Systems), jeff Sponsored by: Nokia
2008-08-13 18:24:22 +00:00
return (0);
}
static void
witness_add_fullgraph(struct sbuf *sb, struct witness *w)
{
int i;
if (w->w_displayed != 0 || (w->w_file == NULL && w->w_line == 0))
return;
w->w_displayed = 1;
WITNESS_INDEX_ASSERT(w->w_index);
for (i = 1; i <= w_max_used_index; i++) {
if (w_rmatrix[w->w_index][i] & WITNESS_PARENT) {
sbuf_printf(sb, "\"%s\",\"%s\"\n", w->w_name,
w_data[i].w_name);
witness_add_fullgraph(sb, &w_data[i]);
}
}
}
/*
* A simple hash function. Takes a key pointer and a key size. If size == 0,
* interprets the key as a string and reads until the null
* terminator. Otherwise, reads the first size bytes. Returns an unsigned 32-bit
* hash value computed from the key.
*/
static uint32_t
witness_hash_djb2(const uint8_t *key, uint32_t size)
{
unsigned int hash = 5381;
int i;
/* hash = hash * 33 + key[i] */
if (size)
for (i = 0; i < size; i++)
hash = ((hash << 5) + hash) + (unsigned int)key[i];
else
for (i = 0; key[i] != 0; i++)
hash = ((hash << 5) + hash) + (unsigned int)key[i];
return (hash);
}
/*
* Initializes the two witness hash tables. Called exactly once from
* witness_initialize().
*/
static void
witness_init_hash_tables(void)
{
int i;
MPASS(witness_cold);
/* Initialize the hash tables. */
for (i = 0; i < WITNESS_HASH_SIZE; i++)
w_hash.wh_array[i] = NULL;
w_hash.wh_size = WITNESS_HASH_SIZE;
w_hash.wh_count = 0;
/* Initialize the lock order data hash. */
w_lofree = NULL;
for (i = 0; i < WITNESS_LO_DATA_COUNT; i++) {
memset(&w_lodata[i], 0, sizeof(w_lodata[i]));
w_lodata[i].wlod_next = w_lofree;
w_lofree = &w_lodata[i];
}
w_lohash.wloh_size = WITNESS_LO_HASH_SIZE;
w_lohash.wloh_count = 0;
for (i = 0; i < WITNESS_LO_HASH_SIZE; i++)
w_lohash.wloh_array[i] = NULL;
}
static struct witness *
witness_hash_get(const char *key)
{
struct witness *w;
uint32_t hash;
MPASS(key != NULL);
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
hash = witness_hash_djb2(key, 0) % w_hash.wh_size;
w = w_hash.wh_array[hash];
while (w != NULL) {
if (strcmp(w->w_name, key) == 0)
goto out;
w = w->w_hash_next;
}
out:
return (w);
}
static void
witness_hash_put(struct witness *w)
{
uint32_t hash;
MPASS(w != NULL);
MPASS(w->w_name != NULL);
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
KASSERT(witness_hash_get(w->w_name) == NULL,
("%s: trying to add a hash entry that already exists!", __func__));
KASSERT(w->w_hash_next == NULL,
("%s: w->w_hash_next != NULL", __func__));
hash = witness_hash_djb2(w->w_name, 0) % w_hash.wh_size;
w->w_hash_next = w_hash.wh_array[hash];
w_hash.wh_array[hash] = w;
w_hash.wh_count++;
}
static struct witness_lock_order_data *
witness_lock_order_get(struct witness *parent, struct witness *child)
{
struct witness_lock_order_data *data = NULL;
struct witness_lock_order_key key;
unsigned int hash;
MPASS(parent != NULL && child != NULL);
key.from = parent->w_index;
key.to = child->w_index;
WITNESS_INDEX_ASSERT(key.from);
WITNESS_INDEX_ASSERT(key.to);
if ((w_rmatrix[parent->w_index][child->w_index]
& WITNESS_LOCK_ORDER_KNOWN) == 0)
goto out;
hash = witness_hash_djb2((const char*)&key,
sizeof(key)) % w_lohash.wloh_size;
data = w_lohash.wloh_array[hash];
while (data != NULL) {
if (witness_lock_order_key_equal(&data->wlod_key, &key))
break;
data = data->wlod_next;
}
out:
return (data);
}
/*
* Verify that parent and child have a known relationship, are not the same,
* and child is actually a child of parent. This is done without w_mtx
* to avoid contention in the common case.
*/
static int
witness_lock_order_check(struct witness *parent, struct witness *child)
{
if (parent != child &&
w_rmatrix[parent->w_index][child->w_index]
& WITNESS_LOCK_ORDER_KNOWN &&
isitmychild(parent, child))
return (1);
return (0);
}
static int
witness_lock_order_add(struct witness *parent, struct witness *child)
{
struct witness_lock_order_data *data = NULL;
struct witness_lock_order_key key;
unsigned int hash;
MPASS(parent != NULL && child != NULL);
key.from = parent->w_index;
key.to = child->w_index;
WITNESS_INDEX_ASSERT(key.from);
WITNESS_INDEX_ASSERT(key.to);
if (w_rmatrix[parent->w_index][child->w_index]
& WITNESS_LOCK_ORDER_KNOWN)
return (1);
hash = witness_hash_djb2((const char*)&key,
sizeof(key)) % w_lohash.wloh_size;
w_rmatrix[parent->w_index][child->w_index] |= WITNESS_LOCK_ORDER_KNOWN;
data = w_lofree;
if (data == NULL)
return (0);
w_lofree = data->wlod_next;
data->wlod_next = w_lohash.wloh_array[hash];
data->wlod_key = key;
w_lohash.wloh_array[hash] = data;
w_lohash.wloh_count++;
stack_zero(&data->wlod_stack);
stack_save(&data->wlod_stack);
return (1);
}
/* Call this whenver the structure of the witness graph changes. */
static void
witness_increment_graph_generation(void)
{
if (witness_cold == 0)
mtx_assert(&w_mtx, MA_OWNED);
w_generation++;
}
#ifdef KDB
static void
_witness_debugger(int cond, const char *msg)
{
if (witness_trace && cond)
kdb_backtrace();
if (witness_kdb && cond)
kdb_enter(KDB_WHY_WITNESS, msg);
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
}
#endif
Reference in New Issue Copy Permalink